Actionable insights — Friday POW!: Gilead Sciences (GILD)

1. Discriminate an accounting event from a business event — the single highest-yield question on an ugly print

1. When a headline loss appears, ask one question before any other: did cash leave, and did the operating business get worse? Those are separable, and the market routinely conflates them.
2. Trace the charge to its accounting rule. Charges forced by a rule — immediate expensing of acquired in-process R&D and upfront payments, goodwill impairment, mark-to-market on a legacy book — are backward-looking bookkeeping about assets you still own.
3. Test the discrimination against other lines in the same release. If guidance was raised and segment revenue grew in the same quarter that produced the loss, the deterioration hypothesis is already falsified by the company's own filing.
4. Ask what the charge bought. Write-downs on acquisitions "reflect long-duration assets whose value accrues over years" — the expense is now, the asset is still on the premises.
5. Estimate the persistence of the optical damage: how many quarters will the number sit in screeners, trailing P/E fields and headlines? That duration is your window, and it is why the mispricing survives.
6. Distinguish this from the inverse case — a business event dressed as an accounting one (a "one-off" that recurs every year, a restructuring charge that never stops). Charges that repeat are operating costs.

• The quarter the charge annualises out of the trailing figures — that is when screeners stop flagging the company and the mechanical selling pressure lifts. And, on the other side, a second large write-down: repeated charges convert the "one-off" reading into an operating-cost reading and break the thesis.

2. Rebuild the earnings power the headline destroyed — margin × revenue, stated as an absolute number

1. Take the adjusted operating margin from a clean quarter (not a trailing average contaminated by the charge) and multiply it by guided annual revenue. One multiplication produces the number the GAAP line is hiding.
2. Prefer operating income to EPS for this: it sits above the financing and one-off lines where the distortion lives.
3. Sanity-check the margin against the peer set — a margin that is an industry outlier needs a structural reason (here, patent-protected pricing on a dominant regimen), otherwise it is being flattered too.
4. Express the result as a plain dollar figure and carry it through the rest of the analysis. "$14 billion of annual adjusted operating income" is harder to argue with than a multiple.
5. Re-strike the valuation multiple on that number, and state both the current multiple and the target multiple you expect once the optics normalise.

• The adjusted operating margin holding near 46-47% in subsequent quarters. Margin erosion would mean the normalised earnings power was overstated and the whole rebuild collapses — this is the number to track, not the GAAP line.

3. Corroborate normalised profit with a cash yield and a debt-paydown test

1. An adjusted number is a claim; free cash flow is the audit. Take forward FCF and divide by the current market capitalisation to get an owner's yield you can compare against a bond.
2. Use the forward estimate, not trailing, when the trailing period contains the distorting charge — and say where the estimate comes from so the reader can discount it.
3. Run the net-debt-to-FCF test: cash minus debt, divided by annual free cash flow. A ratio near 1.0× means the balance sheet is a rounding error and the equity carries no financing risk.
4. Confirm the cash is already being distributed — an established dividend and an active buyback prove the cash flow is real and reaching shareholders rather than sitting in a model.
5. Where the yield is high and the leverage is trivial, the mispricing is almost certainly in the perception, not the finances — which points you back to the optical cause.

• Actual reported free cash flow tracking toward the $12.5B estimate over the next two prints, and whether buyback pace increases once the write-down cycle ends — the clearest management signal that they also see the loss as optical.

4. Reject a bear case on joint probability, not by arguing each leg

1. Enumerate the bear case as discrete, independent risks and name each one plainly — do not blend them into a vague "execution risk."
2. Concede the credible ones explicitly. Saying "we'd be lying if we said it's completely bogus" costs nothing and buys the reader's trust in the parts you do dispute.
3. Ask whether the current price requires all of them to hit, or only one. A discount that only makes sense if two independent bad outcomes occur simultaneously is priced off the joint probability — much lower than either alone.
4. State it as a burden-of-proof claim rather than a forecast: "that is a higher burden of proof than the current price implies." This survives being early and does not require predicting the outcomes.
5. Beware the failure mode: if the risks are correlated (both driven by the same underlying cause), the joint-probability argument is invalid. Check independence before using it.
6. Locate the sell-side dispersion — a wide target range with a named bear and a named bull tells you exactly which risk the market is arguing about, and therefore which datapoint resolves it.

• Biktarvy appearing on a Medicare negotiation list — the single event that resolves the first risk — and any Phase-2/3 failure at the recently acquired programs. Either alone is survivable on this framing; both landing is the thesis-breaker.

5. Distinguish a market expansion from a label footnote — first-line versus later-line

1. For any approval, ask where in the treatment sequence the product now sits. First-line means every newly diagnosed patient is addressable; later-line means only those who have already failed something else.
2. Size the difference in patients, not in indications. Moving from second-line to first-line in the same disease can multiply the treated population several times over without adding a single new indication.
3. Check whether the bears' model uses the old volumes. A consensus built on niche later-line uptake will systematically under-forecast a first-line launch — that gap is the edge.
4. Count approvals across geographies and combinations (US, EU, in-combination with a partner's drug), since each independently widens the addressable base.
5. Re-underwrite the original acquisition price against the current label set. An expensive deal criticised at signing can become defensible purely through label expansion, with no change in the purchase price.

• Reported Trodelvy revenue in the two quarters after each first-line approval — the direct test of whether first-line volumes materialise as claimed — plus additional first-line Phase 3 readouts in other tumour types.

6. Enter on the retest of a prior breakout, and check the multiple hasn't run with the price

1. Find the multi-year breakout — the point where price cleared a range that had capped it for two years or more. That level, once cleared, is the reference for everything after.
2. Accept that you missed the breakout. The question is not whether the move already happened, but whether the valuation still resembles the valuation at the breakout: if earnings grew as fast as the price, the forward multiple is unchanged and the setup is intact.
3. Wait for price to fall back to a long-term trend line (the 200-day moving average), and buy the retest rather than chasing extension. Falling to support is a pause; falling through it is a trend change, and the difference is checkable the same day.
4. Require the fundamental catalyst and the technical setup to coincide. A retest with no re-rating catalyst is just a cheaper price; a catalyst with no retest is a worse entry.
5. Name the invalidation level explicitly, so the position has an exit that does not depend on re-reading the thesis.

• A decisive close below the 200-day moving average, which converts "the pause that refreshes" into a broken trend — and, on the upside, a move back through the 52-week high at $157.29 as confirmation the re-rating toward 15-16× is underway.