Week 20: Earnings vs Cash Flow — Drift, Accruals, and Why FCF Beats EPS

---

1. Why This Is Important

Earnings are an opinion. Cash is a fact. That sentence is older than any of the textbooks that quote it, and it is still the single most useful sentence in fundamental investing.

Reported earnings per share — the number on the wire at 4:01 PM Eastern that the algorithms react to in milliseconds — is the output of dozens of accounting choices. When to recognise revenue. How fast to depreciate. What to capitalise versus expense. How much to reserve for bad debts. Each choice is, individually, defensible. Stacked together, they create enough latitude that the same underlying business can print a $2.10 EPS or a $1.85 EPS depending on the mood of the CFO and the patience of the auditor.

Free cash flow is harder to fake. Cash either showed up in the bank account, or it did not. The cash flow statement is not free of accounting judgement — classification of operating versus investing flows is genuinely ambiguous in places — but the level of cash moves much less than the level of reported earnings under management discretion. That asymmetry is why this week matters:

**Two of the most durable academic anomalies are about earnings

quality, not earnings level.** Bernard & Thomas (1989) documented the post-earnings-announcement drift — stocks with positive surprises continue drifting up for sixty days, stocks with negative surprises continue drifting down. Sloan (1996) documented the accruals anomaly — firms whose earnings are mostly accruals underperform firms whose earnings are mostly cash by roughly seven to ten percentage points per year over the following year. Both anomalies have been replicated dozens of times. Both have weakened since publication. Neither has inverted.

**Almost every spectacular fraud is detectable from earnings

quality.** Enron, WorldCom, Wirecard, Luckin, Valeant. In each case, the gap between reported earnings and operating cash flow was screaming for years before the headline. Reading the gap is not a substitute for an audit. It is a way to never own the companies the auditor will eventually have to disown.

Free cash flow is what valuation actually discounts. The DCF

model in Week 21 discounts FCF, not EPS. A company whose EPS is growing while FCF is shrinking is, by that model, becoming less valuable, not more — regardless of what the analyst note says.

EPS-vs-FCF dispersion widens late in the cycle. When growth

is getting harder to find, earnings management gets more aggressive. Receivables stretch out. Inventory builds. Costs get capitalised. Buybacks juice EPS even as FCF flatlines. The dispersion is itself a regime signal.

This is one of the structural alpha sources in this course: cash-versus-accruals interpretation. It is not a secret. It is not crowded out either, because most of the market still trades on the headline number.

---

2. What You Need to Know

2.1 The Identity: Accruals = Earnings − Cash Flow

Start with arithmetic. For any period:

• Net income = Operating cash flow + Total accruals
• Total accruals = Net income − Operating cash flow

That is not a model. That is the definition. Operating cash flow is already net income with the accruals stripped out — depreciation added back, working-capital changes adjusted, deferred tax adjusted. What is left over after the bridge from net income to OCF is, by construction, accruals.

Accruals are not bad. They exist for a reason. Revenue earned but not yet collected — receivables — is a real asset. Inventory built ahead of demand is a real economic activity. Depreciation, the largest single accrual at most industrial firms, is an honest attempt to match an asset's cost to the periods that asset is producing.

The problem is that accruals are also the lever managers reach for when the quarter is short. Every line on the bridge from net income to OCF is a discretionary judgement, and every one of them, when pushed hard enough, increases reported earnings without increasing real cash.

Two ways to compute accruals for a stock screen:

• Balance-sheet method (Sloan): change in non-cash current

assets, minus change in current liabilities (excl. short-term debt), minus depreciation. Scale by average total assets. Sort firms into quintiles annually. Long Q1 (lowest), short Q5 (highest).

• Cash-flow method: simply (Net income − Operating cash flow) /

average total assets. Equivalent in spirit, slightly cleaner since 2002 when SFAS rules tightened.

The accruals anomaly is shown below:

The Q1-minus-Q5 spread is the academic estimate of the alpha. In a real portfolio, half of it gets eaten by transaction costs, capacity limits, and short-borrow fees on the Q5 leg. But even after that, this is one of the best risk-adjusted long-short signals ever documented in equity markets.

2.2 The Post-Earnings-Announcement Drift (PEAD)

The PEAD is the closest cousin to the accruals anomaly. Bernard & Thomas (1989) ranked every quarterly earnings announcement by standardised unexpected earnings (SUE) — the surprise relative to analyst consensus, divided by the historical standard deviation of that surprise. Top quintile vs bottom quintile, post-announcement returns over sixty trading days, averaged over thousands of announcements: the result is stable enough to be boring.

The drift is incomplete absorption of news. The market reacts on Day 0 — the day of the release and the conference call — but the reaction is, on average, only about two thirds of what it should be. The remaining third leaks out over the next sixty days as analyst revisions catch up, as the next quarter's pre-announcement narrows expectations, and as slower investors finally trade on what fast ones already saw.

Why does this not get arbitraged away? Three reasons that show up across all of the persistent earnings anomalies:

• Limits to arbitrage. Borrow on the Q5 leg is genuinely

expensive. Many of the most accrual-heavy firms are small-cap or mid-cap with shallow short-borrow markets. The frictions eat the alpha for funds that need to size up.

• Career risk. A long-short manager whose Q1 leg has a six-month

drawdown will be redeemed before the academic mean-reversion arrives. Fading the crowd is correct on average and dangerous in any individual year.

• The signal is slow. PEAD is a sixty-day phenomenon. The

accruals anomaly is a twelve-month phenomenon. Most of the market is trading on a one-day phenomenon: the print versus the consensus. The slow signals are not crowded by the fast traders.

This connects to the two sides of price discovery: momentum and mean-reversion. PEAD is fundamentally a momentum effect — the stock that surprised up keeps drifting up — because the information is travelling through the market slower than the price. The accruals anomaly is a mean-reversion effect — earnings inflated by accruals revert toward cash over the following year. They sit on opposite sides of the same coin.

2.3 Free Cash Flow, Carefully Defined

FCF gets quoted three different ways in the wild. You need to know which one you are reading.

• FCF to the firm (FCFF) = OCF − Capex. The plain definition.

The cash left over after maintaining and growing the asset base, before any return to debt or equity holders. This is the number the DCF model wants.

• **FCF to equity (FCFE) = FCFF − Net debt repayment + Net debt

issuance.* The cash left over for equity* holders specifically. Used in dividend-discount frameworks.

• "Adjusted FCF" (whatever management says it is). Sometimes

adds back stock-based compensation. Sometimes excludes acquisition-related capex. Sometimes excludes "one-time" working-capital movements that have shown up four years in a row. Treat all adjusted FCF the way you treat adjusted EBITDA: with one eyebrow permanently raised.

For screening, use FCFF. Use a multi-year average — three to five years — because single-year FCF is noisy from working-capital and capex timing. The Apple FY2014-FY2024 chart from Week 8 is the canonical example: EPS and FCF/share track within pennies most years, and where they diverge (FY2020 working-capital release; FY2024 EU tax charge), the divergence has a clean accounting explanation.

The interactive lab below lets you flip between six representative companies and see the EPS-vs-FCF profile and the per-year accrual:

Open: Earnings vs Cash Flow Lab

The pattern matters more than any single year. Apple's FCF runs above EPS most years — capital-light, working-capital favourable. Coca-Cola's two lines hug each other within pennies — steady-state brand-rent business. Microsoft's FCF lagged EPS in FY2023-2024 not because of manipulation but because of the AI capex bulge — that is an investment story, not a quality story. Amazon shows the textbook cash-light-but-EPS-light case where huge non-cash D&A makes earnings look thin while the cash story (post-2022 capex moderation) is what actually matters. GE shows the opposite — multi-year restructuring charges depress EPS far below the underlying cash. JPM is a bank; "FCF" for a bank is conceptually meaningless, and you need ROTCE instead (covered in Week 19).

2.4 Earnings Quality Red Flags — The Five That Matter

A short list. Most quality work in the wild is a variation of these five.

1. Days sales outstanding (DSO) rising faster than revenue. DSO = receivables / (revenue / 365). If revenue grew 12% and DSO grew 30%, the company is "selling" to customers who are not yet paying. Revenue is being recognised earlier in the cycle, or to customers who will not actually pay. Aggressive DSO is the most common precursor to a revenue restatement.

2. Days inventory outstanding (DIO) rising faster than revenue. DIO = inventory / (COGS / 365). Inventory builds either because the company is building ahead of expected demand (sometimes legitimate) or because the inventory the company already produced is not selling (usually the bad version). Either way, it is a use of cash that does not show up on the income statement until the markdown finally hits COGS.

3. Capitalisation of costs that should be expensed. R&D under US GAAP is mostly expensed; under IFRS, much of it can be capitalised. Software development costs, content production costs, acquisition-related "integration" costs — the more cost a firm moves to the balance sheet, the higher today's earnings and the lower tomorrow's. WorldCom's fraud was, mechanically, this exact move on $11B of network operating expense.

4. Recurring "one-time" charges. A company that takes a large restructuring charge in three out of four years is not restructuring; it is telling you the underlying earnings power is lower than the "adjusted" headline. The charges are real cost. They just keep getting flagged as not-real-cost.

5. The accrual ratio itself. (Net income − OCF) / Net income, or scaled by average assets, year after year. A persistent positive gap is a quality concern. A persistent negative gap (FCF > EPS) is the opposite — the signature of a capital-light business with non-cash charges dragging on earnings.

None of these are deterministic. Each has a legitimate version. But in a screen, the firms that flunk three of the five tests have a materially worse forward return distribution than the firms that pass them all.

2.5 Late-Cycle Dispersion — Why This Week Matters in 2026

Through 2024 and into 2025, the EPS-vs-FCF dispersion in the S&P 500 widened. Reported earnings continued to grow at 8-10% per year while operating cash flow growth slowed to 3-4%. Some of that gap is real — AI capex is genuinely investment, not manipulation. Some of it is the late-cycle pattern — DSO and DIO drifting up, "adjusted" metrics drifting further from GAAP, share buybacks juicing EPS while FCF flatlines. The cycle is not the same every time. The gap behaves roughly the same every time.

The contrarian read: the firms whose FCF is keeping pace with EPS in this environment are getting a quality premium that the headline EPS multiple does not yet reflect. The firms whose EPS-FCF gap has widened for three years running are tomorrow's accrual reversion candidates. This is why "look at cash, not earnings" is a structural alpha source, not a tactical one. It works in every regime. It works hardest when the rest of the market has stopped checking.

---

3. Common Misconceptions

"FCF is always better than EPS." Wrong. FCF is more useful

for quality questions; EPS is more useful for profitability comparisons within a stable industry. A capex-heavy firm in year-three of a four-year build will show ugly FCF and fine EPS. That is not a quality problem. It is a timing problem.

"Accruals are accounting fraud." Wrong. Accruals are

accounting. Without them you would have cash accounting, which is useless for businesses with multi-period contracts, inventory, or long-lived assets. The question is not whether accruals exist; it is whether they are growing faster than economic reality.

"PEAD has been arbitraged away." Mostly wrong. The drift has

shrunk — Bernard & Thomas reported a 60-day spread of 4-5% in the 1980s; modern estimates are 2-3% for the top vs bottom quintile. But the asymmetry has not flipped, and the drift is still detectable in every replication through 2024.

"A negative FCF year is always bad." Wrong. Amazon ran

negative or near-zero FCF for most of its first decade as a public company while building infrastructure that is now generating tens of billions per year. Negative FCF in service of a high-return investment programme is great. Negative FCF because the working capital is hemorrhaging is fatal. Look at the cause.

"EPS beats are what matter." Misleading. The EPS print

versus consensus matters in the next 24 hours. The PEAD says it also matters in the next 60 days. But the multi-year story — quality of those earnings, conversion to cash — is what compounds the stock price over a 5-10-year hold.

**"Stock-based compensation is non-cash so I should add it

back."** Wrong, mostly. SBC is a real cost — the firm is giving away part of the company. The cash that would have been spent on payroll is instead being raised by issuing new shares, which dilutes you. The right adjustment is to subtract SBC from FCF, not add it back, and many "adjusted FCF" presentations get this exactly backwards.

"Revenue can't be faked." Wrong. Revenue is the most

audit-tested line, but channel-stuffing, bill-and-hold, round-trip transactions, and aggressive percentage-of-completion accounting have all been used to inflate revenue in real cases. The DSO check exists precisely because revenue can be pulled forward.

"If a company beats earnings, it's a good investment."

Wrong. The PEAD says you get some drift after a beat, on average, in excess of the immediate reaction. But the average masks dispersion — beats that come with deteriorating cash flow tend to mean-revert hard. Beat plus rising accruals is a worse profile than in-line plus stable accruals.

"FCF yield is the new P/E." Useful, not magical. FCF yield

(FCF / market cap) is a cleaner valuation metric for mature non-bank businesses than P/E. For high-growth firms with negative FCF, it is meaningless. For banks and insurers, it is meaningless. For asset-heavy capex-cycle firms, single-year FCF yield is noisy — average over five years.

**"Forensic accounting is for activists, not retail

investors."** Wrong. The five red flags above are computable from the standard 10-K filing in 30 minutes per company. They will not catch a sophisticated fraud. They will catch the obvious ones, and they will tell you which 10-Ks deserve another hour of your time.

---

4. Q&A Section

**Q1: How quickly can I screen for accruals on US-listed equities?** A: With a paid data feed (Compustat, FactSet, S&P Capital IQ) the ratio is one column and the quintile sort is one query. With free sources, pull (Net Income − OCF) from the 10-K cash flow statement for each name in your watchlist; scale by average total assets. A twenty-name watchlist takes an hour the first time and ten minutes per quarter to refresh.

Q2: Should I short the high-accrual names? A: Probably not as a retail investor. Short borrow on small- and mid-cap names is expensive, mark-to-market drawdowns can be sharp, and short squeezes happen even on terminal cases. The cleaner implementation is to screen those names out of your long book and concentrate on the long-cash quality side. That captures most of the asymmetry without the operational complexity — long-cash quality goes in the core tranche; the short side, if you run it at all, sits in a specialty tranche.

Q3: Why does PEAD drift continue for 60 days and not longer? A: Around the 60-day mark the next quarter's pre-announcement window opens and forward expectations start being discounted into the price. The "old" surprise is no longer the marginal news. Empirically the drift fades into the next earnings cycle, not after a fixed calendar window.

Q4: Is the accruals anomaly the same as the quality factor? A: Related, not identical. The "quality" factor as MSCI and others construct it usually combines accruals, gross profitability, leverage, and earnings stability. Accruals is one input. The pure accruals anomaly is more concentrated than the blended quality factor and historically had a bigger Sharpe.

Q5: What about non-US markets? A: The accruals anomaly has been replicated in most developed markets — UK, Japan, Australia, Western Europe — at varying strengths. In emerging markets the data quality is weaker and the shorting frictions worse. Our investable universe in this course is US-listed equities anyway, so this is not a real constraint for us.

Q6: How do I use FCF yield in valuation? A: For mature non-bank firms, an FCF yield (5-year average FCF over current market cap) above the long Treasury yield plus a sensible equity premium is a reasonable starting screen. So if 10-year Treasuries are at 4.5% and you want a 4-5% equity premium, you are looking for FCF yield above 8.5-9.5%. Below that you are paying for expected growth; above that you are being paid to wait.

Q7: Why did Microsoft's FCF lag EPS in 2024? A: Capex. Microsoft is spending $50-60B per year on AI data centres, chips, and cooling infrastructure. That capex is real and is flattening FCF even though OCF and EPS are at all-time highs. Whether the AI capex earns its cost of capital is the real question — the FCF gap itself is informative but not by itself a quality problem.

**Q8: What's the difference between "earnings management" and "earnings manipulation"?** A: Mostly intent and degree. Both shift earnings between periods using legal accounting choices. Management is done quietly to smooth quarterly volatility — almost every public company does some of it. Manipulation is done aggressively to hit a target the company would otherwise miss, and shades into fraud when the choices stop having a defensible accounting basis. The earnings-quality red flags catch both.

Q9: How does this lesson interact with valuation (Week 21)? A: Directly. The DCF in Week 21 discounts free cash flow, which means everything you learned this week about the quality of FCF feeds straight in. Two firms with identical projected FCF deserve different multiples if one has a clean accruals profile and the other does not. The market does not always price that gap; you can.

**Q10: What about software companies with deferred revenue and SBC?** A: Two complications, opposite directions. Deferred revenue (cash received for services not yet rendered) inflates OCF relative to EPS; that is a legitimate tilt, not a quality issue, as long as the underlying contracts are renewable. SBC, as discussed in misconception #6, is a real cost and should be subtracted from FCF to get the true equity-holder cash. The net effect varies by firm — modern SaaS often shows very high OCF and middling SBC-adjusted FCF.

Q11: Is there a "bad" version of FCF > EPS? A: Rarely, but yes. A firm that is shrinking — drawing down working capital and not replacing assets — can post FCF above EPS for a year or two while the business is being liquidated. The clue is declining revenue and declining capex. The healthy version of FCF > EPS is a stable or growing capital-light business. The unhealthy version is a melting ice cube monetising its inventory and receivables one last time.

Q12: How much weight should I put on a single quarter? A: Less than the headlines do. PEAD says the print does matter for the next 60 days. But the 12-month forward return is much more correlated with multi-year accrual quality and cash-flow conversion than with any single quarter's surprise. Quarter-to-quarter volatility is mostly noise; the trend in EPS-vs-FCF dispersion is signal.