Verdict: Keep

BKNG is 0.77% of the selectable universe. Factor profile is borderline — 68.5% idio variance (below 75% target), MTUM beta -0.20, trailing alpha -21.1% annualized — but every metric is better than the weakest confirmed remove. The anti-momentum is half the magnitude of MAR (-0.57) or CSX (-0.46). The trailing alpha is dominated by the Feb 2026 OTA sector selloff, not structural decay.

Deep-dive into the 10-K, earnings call, balance sheet, options market, and cross-corpus travel transcripts found no actionable mispricing in either direction for the 15-week window. The structural concerns (CAC inflation, earnings quality divergence, financial engineering) are real but 12-24 month theses. The 15-week setup (oversold, stock split April 6, Q1 earnings beat pattern, summer seasonality) favors modest outperformance. At 0.77% weight, max plausible drag from a wrong keep is 8 bps. Cannot prove structural underperformance. Keep.

Q1 earnings April 28 is the decision gate. Room nights <5% + marketing costs up again = revisit removal.

Factor Profile

250-day regression against SPY + MTUM:

Alpha (ann): -21.1%. Idio vol: 26.6%. Total vol: 32.1%. R-squared: 31.5%.

Comparison to confirmed removes:

Pattern for factor-based removes: idio <62%, MTUM beta < -0.27. BKNG doesn't fit. The trailing alpha is misleading — the -21.1% annualized is dominated by the Feb 2026 OTA sector selloff (BKNG -20%, EXPE -27%, TRIP -30% simultaneously on AI disruption fear). The regression captures this as "idiosyncratic alpha" but it was a thematic factor re-rating. Recent 1M relative strength: +7.1pp vs QQQ. Momentum is inflecting.

The Income Statement Nobody's Reading

Revenue grew 13%. Operating income grew 17%. Net income fell 8%. Everything between operating income and net income destroyed the year.

The below-the-line damage breaks down to three drivers:

1. Net interest cost: -$515M deterioration. Interest expense up $322M (more debt), interest income down $193M. Structural. $18.7B gross debt ($1.9B short-term + $16.9B long-term) costs $1.6B/year to service. They issued $3.7B new debt in 2025 while redeeming $5.0B — net refi at higher rates. This doesn't reverse. (10-K lines 3103-3109)

2. Other income: -$1,215M swing. Driven by unrealized FX losses on Euro-denominated debt ($1,428M per cash flow statement adjustments). BKNG maintains large Euro-denominated note issuances as a natural hedge for European revenue, but when EUR/USD moves, GAAP earnings swing $1B+ in either direction. This is volatile, not structurally negative — but it IS a consequence of the debt architecture. In a 15-week window where EUR/USD could move 3-5%, Q1 GAAP numbers become unpredictable regardless of operational performance. (10-K line 4163-4169)

3. KAYAK impairment: $457M (new line item). $180M goodwill + $277M intangible assets. The 10-K says: "reduction in forecasted cash flows for KAYAK, reflecting its meta-search business being impacted by expected increases in customer acquisition costs." One-time charge, structural cause. This is management's Tier 1 admission, under auditor scrutiny, that AI is killing the meta-search referral layer. The charge won't recur (you only write down KAYAK once), but the DRIVER — AI-mediated CAC inflation — persists. (10-K lines 2505, KAYAK impairment note)

The GAAP/Adjusted gap: $61.50/share, widened 5.1x YoY. Adjusted EPS adds back SBC ($617M), impairment ($457M), transformation costs ($205M), and various below-line items. Steenbergen himself says on the Q4 call (line 57): "we believe stock-based compensation is a real operating cost" — credit to management for including SBC in adjusted numbers. But the other add-backs are material. Strip just the impairment and transformation costs, and "adjusted" EPS is closer to $208, implying forward P/E of ≈15-16x, not the headline 13.5x.

The Balance Sheet Machine

Negative equity deepened $1.6B. They bought back $6.4B of stock while earning $5.4B in net income and paying $1.3B in dividends. The buyback machine runs faster than the earnings engine — that's the definition of financial engineering. Average repurchase prices: 2023 ≈$2,800/share, 2024 ≈$3,842/share, 2025 ≈$4,960/share. Management is currently underwater on all 2025 buybacks. At current $4,214, they overpaid by ≈15%.

Remaining authorization: $21.8B. At today's price that buys ≈5.2M shares, or ≈16% of outstanding over 3-4 years. That's a structural bid under the stock — but also further deepening negative equity. Investment-grade credit (Baa1/BBB+) and $9.1B FCF mean this is aggressive, not distressed. But there's no balance sheet cushion if travel demand cracks. (10-K, stockholders' equity table, lines 4107-4140)

Market Consensus: Three Layers

Layer 1: Sell-Side

Consensus median PT: ≈$5,400. 30% upside. Q1 EPS estimate: $27.47 (Apr 28). Four of six PTs cluster between $5,400-$6,250 — that's the herd. The narrative: AI fear overdone, 13.5x is a steal, summer bounce.

Layer 2: Options Market

May 15 (48 DTE, spans Q1 earnings): P/C OI ratio 0.86 — neutral. Market is hedged around earnings, not directional. Max pain $4,300 — pinned near current levels.

July 17 (111 DTE, end of 15-week window): P/C OI ratio 0.63 — bullish. Calls 1.6x puts. P/C volume also 0.62 confirming bullish flow. Heaviest put OI: $3,500 (156 contracts, the tail hedge at -17%). Heaviest call OI: $6,000-$6,500 (scattered lottery tickets at +42-54%). Max pain $4,300.

Read: Neutral through earnings, bullish through summer. Put protection concentrated at one strike ($3,500) — that's the downside the options market is pricing. Call positioning scattered and aspirational. The real expectation: $4,000-$4,800 range through July, slight upward bias.

Layer 3: Adjacent Names (Cross-Corpus Evidence)

Every travel-adjacent Q4 2025 transcript confirms healthy demand:

• GBTG (corporate travel): "Corporate travel demand continued to accelerate in Q4" — but added "short-term negative impact U.S." New qualifier.
• VIK (Viking cruises): 86% booked for 2026, +7% YoY. $6.0B advanced bookings, +13%.
• Hotel REITs (PEB, RLJ, CLDT): All segments recovering. SF RevPAR +37.9%. World Cup 2026 tailwind.
• NCLH (Norwegian Cruise): "Enter 2026 slightly below optimal booking range" — lone soft signal.
• APLE (Apple Hospitality REIT): OTA channel mix UP 110bps to 14% of room nights. Hotels sending MORE business to OTAs, not less. This directly contradicts the "AI kills OTAs" fear.
• TCOM (Trip.com/Ctrip): "Robust travel demand" in China. Silver economy driving growth.
• Airlines (ICAGY/British Airways): "Strong premium cabin business travel demand, particularly U.S."

Adjacent verdict: Travel demand broad and healthy. No cracks except NCLH cruise softness. The APLE data point — OTA channel share growing at hotels — is the strongest counter to the disintermediation thesis and appears to be completely unnoticed by sell-side coverage of BKNG.

Three Mispricing Candidates

1. The "13.5x Is Cheap" Illusion — Mild, Favors Bears Long-Term

Forward P/E 13.5x vs 10-year average 22-25x is the consensus anchor. But: (a) SBC is already included in BKNG's adjusted EPS, so this comparison is fair on that axis; (b) however, the other add-backs ($457M impairment + $205M transformation + below-line items) inflate adjusted EPS by ≈$20/share; (c) stripping just impairment and transformation costs, forward P/E is 15-16x, not 13.5x; (d) below-the-line structural costs (net interest -$515M deterioration, Euro debt FX volatility) make the GAAP earnings path more volatile than the adjusted path suggests.

Mispricing: ≈2 P/E turns. Market thinks 13.5x for a 22-25x business. More like 15-16x for a business whose below-the-line costs are structurally higher. Not egregious. Not actionable in 15 weeks.

2. CAC Inflation as Structural, Not Cyclical — Real, Not Yet Priced

The 10-K (line 2320-2321): "our average ROI was down slightly year-over-year driven by changes in paid traffic mix and increased social media spend."

The earnings call (Steenbergen, line 94): "performance marketing, in attractive ROIs."

The 10-K is what management stands behind under SEC liability. The call is spin. Both can technically be true (individual channels positive, blended average declining as mix shifts to lower-ROI channels). But the direction matters.

Marketing: $8.2B (+12.5%), up 24bps as % of gross bookings. Direct channel: mid-60% (up YoY). So direct is growing AND paid marketing is getting more expensive simultaneously. The paid channel is getting less efficient even as the free channel scales. The 10-K risk factor (lines 786-788) explicitly warns: "placement of AI-generated content by these platforms could... increase customer acquisition costs."

The KAYAK impairment is the scoping mechanism. Management wrote down the meta-search layer citing "expected increases in customer acquisition costs" — while simultaneously guiding core business above long-term algorithm. They're telling you: meta-search dies from AI, full-stack OTA transaction layer survives. The question consensus is ignoring: if CAC inflation is real for meta-search, it eventually reaches the OTA layer too, just at a slower rate. If marketing efficiency degrades 50-100bps/year as % of gross bookings, that's $650M-$1.3B annual EBITDA drag.

Timing: Slow burn. Q1 will provide one data point (marketing costs up/down), not resolution. 4-6 quarter thesis, not 15 weeks.

3. The FX Volatility Bomb — Not Mispriced, Misunderstood

"Other income" swung -$1,215M YoY, mostly unrealized FX on Euro-denominated debt. The market correctly treats this as GAAP noise. But the volatility is underappreciated: BKNG's quarterly GAAP earnings can swing $500M+ on EUR/USD moves alone. This makes Q1 GAAP numbers unpredictable regardless of operational performance. Not a mispricing — a source of potential confusion around April 28 that could create post-earnings volatility in either direction.

AI Disruption: Scoped, Not Solved

De-risking Evidence (Near-Term)

OpenAI Instant Checkout failure (March 2026). Users researched trips in ChatGPT but near-zero transaction completion. "In travel, the transaction is never just the transaction" — pricing volatility, fare rules, ancillaries, cancellations, payment risk, post-booking servicing. OpenAI pivoted to pushing transactions to OTA apps within ChatGPT. BKNG +8% on March 5. This validates the transaction layer moat. (Multiple independent sources: Skift, X/Twitter, Tier 3-4. LR 1.5-2.0)

Fogel's defense (transcript lines 125-139). Not hand-waving. Merchant-of-record complexity: 100+ payment methods, 50+ currencies, 4.4M properties, "several thousand" partner services people on the ground, regulatory burden across 200+ countries (DSA, DNA, DFA, AIA, P2B, DAC7). The LENGTH of the defense is the tell — 40 lines explaining why LLMs can't do what OTAs do means he's thought about the threat seriously. His Google analogy frames LLMs as the new top-of-funnel, OTAs as the transaction layer. The OTA playbook against Google has worked for 20 years.

Structural Risk Evidence (Long-Term)

KAYAK impairment ($457M). Management's Tier 1 admission that AI kills the meta-search referral layer. Scoping mechanism: what's impaired (referral) vs what's not (transaction). (10-K, Tier 1, LR 0.8 for core OTA)

10-K risk factor (lines 686-698). "AI agents may further evolve into full-service booking platforms, increasing competitive pressure and disintermediating OTCs." Filed under SEC liability. Management believes this is possible. (10-K, Tier 1, LR 0.8)

Marketing ROI decline (10-K line 2321). Blended ROI down slightly despite growing direct channel. If AI shifts customer discovery upstream, the marginal non-direct customer costs more to acquire. (10-K, Tier 1, LR 0.8)

30% of travelers using ChatGPT for trip planning (up 124% YoY from 13%). Discovery is shifting upstream. But discovery ≠ transaction — the OpenAI failure confirmed that gap. (Tier 3, LR 1.0 neutral — direction clear, magnitude unclear)

Net AI assessment: The existential threat (LLMs handling transactions directly) has been tested and failed. The slow-burn threat (CAC inflation, discovery shift) is real but multi-year. For 15 weeks: net de-risked. The KAYAK impairment scoped the damage precisely.

Why We Have No Edge

$136B market cap. 31 analysts. RSI 33 at 22% of 52-week range. Short interest 3.0%. Everyone who can spell "OTA" has the Morgan Stanley upgrade note memorized.

What we see that the market is slow to process:

• The 10-K vs earnings call discrepancy on marketing ROI (requires reading both documents side by side; sell-side notes summarize the call)
• The KAYAK impairment as a scoping mechanism (most coverage treats the impairment as a negative; it's actually informative about what's NOT impaired)
• The APLE data point on OTA channel share growth (buried in a hotel REIT transcript nobody reads for BKNG intelligence)

What we definitely don't have edge on: Valuation (every analyst has a DCF), AI disruption narrative (every tech publication covers it), Q1 room night trajectory (STR data available to institutional desks), options positioning (visible to any terminal).

Counterparty analysis: If we removed BKNG and it bounced, our counterparty would be bottom-fishers buying at 13.5x forward with AI fear de-risking, summer seasonality ahead, and a $21.8B buyback authorization. That's an informed buyer. Our edge over that buyer: zero.

Catalyst Calendar (15-Week Window)

Invalidation Triggers (Would Change to FILTER)

• Q1 room nights <5% — Deceleration hardening, consumer thesis deteriorates
• Q1 marketing as % of gross bookings up again — Two consecutive quarters of rising marketing intensity = structural CAC inflation, not tactical
• Direct channel mix declining — AI eroding the loyalty/direct moat
• Another AI transaction platform launch — Google Overviews checkout, Perplexity/Claude travel booking. OpenAI failed once; doesn't mean the next attempt fails.
• Macro recession evidence — BKNG has no B2B hedge (unlike EXPE at 33% B2B). Purely discretionary consumer.

None present in current data.

Sources: BKNG 10-K (SEC, 2026-02-18) — primary [Tier 1]. BKNG Q4 2025 earnings call (2026-02-18) [Tier 2]. Factor regression: 250d trailing SPY + MTUM [Quant]. yfinance market data (2026-03-27) [Tier 2]. Options chain May/July (yfinance) [Tier 2]. Cross-corpus transcript search: GBTG, VIK, NCLH, PEB, RLJ, CLDT, APLE, TCOM, ICAGY Q4 2025 calls (5,762 transcripts searched) [Tier 2]. OpenAI checkout walkback — Skift, X/Twitter multiple independent sources [Tier 3-4]. Morgan Stanley upgrade (2026-02-23) [Tier 3]. Confirmed remove regressions: MAR, CSX, CRWD, SBUX [Quant].