Why it matters: 단기 전략 백테스트는 실제 fill logic을 잘못 모델링하면 edge를 과대평가한다. 특히 passive limit order는 싸게 사는 도구가 아니라 adverse selection을 받아들이는 보험 판매와 비슷해질 수 있다. Henry가 execution cost를 과소평가한 채 leverage를 올리는 것을 막는 자료다.
Raw excerpt / abstract: The paper studies fill probabilities and adverse fills in market-making simulations on ES, NQ, CL, and ZN. It says many simulations treat price processes and market orders independently, which can largely inflate short-term strategy performance. The authors distinguish adverse fills from non-adverse fills and show that realistic fill probabilities and adverse fills make results closer to reality.
Cold read: backtest PnL의 상당 부분은 alpha가 아니라 fill assumption일 수 있다. 시장은 좋은 가격만 주지 않고, 특히 가격이 내 limit order를 통과할 때 나쁜 fill을 강제로 준다.
Trading insight: passive entry를 쓰더라도 fill 이후 mark-to-market drift를 따로 측정해야 한다. 전략 평가에는 win rate보다 adverse fill rate, queue priority, non-adverse fill probability가 먼저 들어가야 한다.
Change sensor: adverse fill 비중, post-only fill 이후 1초와 5초 drift, queue depth 대비 내 주문 크기.
Follow-up question: Henry의 BTC, MSTR, BMNR 주문에서 limit fill 이후 1분 수익률은 market order보다 실제로 유리한가?
[01:03 KST] Aggregating Orderbook Depth to Create Liquidity Metrics
Why it matters: crypto liquidity는 거래소별로 분산되어 있고 quote asset도 USD, USDC, USDT 등으로 나뉜다. 단일 거래소 depth만 보고 큰 포지션을 잡으면 실제 exit capacity를 착각할 수 있다. 이 자료는 BTC depth를 여러 venue와 stablecoin pair로 합산하는 실무적 방법을 보여준다.
Raw excerpt / abstract: Coin Metrics says a liquid order book is necessary for a mature financial asset and defines the ability to quickly enter and exit large positions with small price impact or slippage as desirable for institutions. The tutorial aggregates BTC market depth within plus or minus 1 percent across USD and stablecoin pairs.
Cold read: visible liquidity는 venue별 조각의 합이며, stress에서는 그 합이 동시에 사라질 수 있다. depth metric은 평시 capacity가 아니라 stress haircut 전 원재료다.
Trading insight: BTC sizing에는 portfolio conviction뿐 아니라 1 percent depth 대비 position size ratio가 필요하다. ETF와 spot venue가 연결되어도 execution window에는 venue fragmentation이 남는다.
Change sensor: plus or minus 1 percent BTC depth, stablecoin quote별 depth share, top venue concentration.
Follow-up question: Henry가 보유한 BTC-related exposure의 하루 exit notional은 aggregate BTC 1 percent depth의 몇 percent인가?
Henry Risk Link: liquidity | narrative | execution
Why it matters: BTC ETF flow는 실시간 demand가 아니라 settlement와 create redeem workflow의 산물이다. Henry가 social media에서 당일 flow headline을 보고 FOMO를 느낄 때, 그것이 이미 전일 trading activity의 후행 지표일 수 있음을 상기시킨다.
Raw excerpt / abstract: NYDIG explains that the key distinction between trade date T and T plus 1 is timing. ETF securities and spot bitcoin trade on T, while movement of funds and creation or redemption of shares takes place on T plus 1. It finds R squared above 70 percent between ETF turnover on T and reported fund flows on T plus 1, and estimates flows around 29 percent of prior day turnover plus intercept. It also notes weekend funding costs reduced Monday reported creations and redemptions.
Cold read: ETF flow는 demand thermometer가 아니라 market maker financing and settlement plumbing의 output이다. on-chain custody movement도 대부분 backward-looking settlement다.
Trading insight: ETF flow headline을 entry trigger로 쓰면 늦을 수 있다. 더 나은 sensor는 prior day turnover, AP activity window, funding cost calendar, weekend settlement friction이다.
Why it matters: Treasury liquidity는 모든 risk asset의 discount rate와 collateral system의 기반이다. 이 자료는 bid-ask spread, order book depth, price impact를 같이 봐야 liquidity regime을 판단할 수 있음을 보여준다. BTC와 MSTR 같은 high beta asset도 결국 USD collateral stress에 영향을 받는다.
Raw excerpt / abstract: NY Fed defines market liquidity as the cost of quickly converting an asset into cash. It tracks bid-ask spread, order book depth, and price impact for on-the-run two-, five-, and ten-year notes. Liquidity worsened sharply after the April 2025 tariff announcement, then improved quickly and reached the best level since 2021 by early 2026, though geopolitical uncertainty after the sample period warrants continued monitoring.
Cold read: Treasury market은 평시에는 깊어 보이지만 policy shock에는 volatility가 liquidity를 즉시 갉아먹는다. liquidity는 상태가 아니라 volatility의 함수다.
Trading insight: macro shock 때 crypto drawdown을 설명하려면 BTC-specific narrative보다 Treasury price impact와 depth deterioration을 먼저 확인해야 한다. cross-asset de-risking의 시작점은 Treasury liquidity다.
Change sensor: Treasury realized volatility, on-the-run depth, price impact per USD 100 million net order flow.
Why it matters: ON RRP drawdown은 reserve drain을 늦추는 buffer로 해석되어 왔다. 그러나 buffer의 소진 속도와 repo market absorption capacity는 다르다. Henry가 liquidity easing을 단순 bullish로 해석하지 않도록, TGA, bills, private repo, SLR constraint를 같이 보게 한다.
Raw excerpt / abstract: Kansas City Fed says balance sheet reduction must show up through TGA, reserves, or ON RRP. ON RRP fell rapidly as Treasury rebuilt the TGA and money market funds bought bills, but further declines may slow as TGA nears projected levels. Private repo absorption may be limited by market size and bank balance-sheet constraints including SLR.
Cold read: ON RRP는 무한한 liquidity reservoir가 아니라 money market plumbing의 한 칸이다. TGA rebuild가 끝나면 marginal liquidity pressure는 reserves와 repo balance sheet로 이동한다.
Trading insight: liquidity regime 판단은 ON RRP level 하나로 하면 안 된다. bills yield premium, repo demand, reserve level, SLR binding을 함께 봐야 risk budget을 정할 수 있다.
Change sensor: ON RRP balance, TGA path, one-month bill premium over ON RRP, private repo share.
Follow-up question: Henry의 high beta exposure에는 ON RRP 잔고보다 reserve scarcity proxy가 더 유효한가?
Why it matters: current macro liquidity는 narrative보다 live plumbing data에서 먼저 보인다. NY Fed dashboard는 reverse repo, repo, SOFR, BGCR, TGCR, SOMA holdings, primary dealer statistics를 한 화면에서 제공한다. Henry의 daily risk throttle에 넣기 좋은 primary source다.
Raw excerpt / abstract: The dashboard shows June 2026 market operations and reference rates. Reverse Repo Operations on June 5, 2026 accepted USD 0.761 billion at 3.50 percent. Repo Operations on June 8, 2026 accepted USD 0.000 billion at 3.75 percent. SOFR for June 5, 2026 was 3.63 percent on USD 3,131 billion volume. SOMA total holdings were USD 6,331,014,782.3 thousand on June 4, 2026.
Cold read: liquidity stress는 뉴스가 아니라 funding rates와 facility usage의 미세한 변화로 먼저 나타난다. repo take-up이 0이어도 quarter-end, Treasury issuance, dealer balance sheet가 바뀌면 regime은 빠르게 전환될 수 있다.
Trading insight: BTC와 MSTR sizing을 늘리기 전 SOFR volume, repo usage, reverse repo usage, SOMA change를 pre-trade checklist에 넣어야 한다. risk asset bet은 funding plumbing이 조용할 때 더 좋은 R-ratio를 가진다.
Why it matters: Kelly는 장기 성장률을 극대화하지만 실제 trader에게 견디기 어려운 drawdown을 만든다. risk-constrained Kelly는 minimum wealth drawdown probability를 constraint로 넣어 equity curve를 부드럽게 만든다. Henry의 대형 베팅은 full Kelly가 아니라 drawdown survival constraint를 먼저 통과해야 한다.
Raw excerpt / abstract: QuantInsti explains that classical Kelly can cause long-lasting big drawdowns. Busseti, Ryu, and Boyd propose a risk-constrained Kelly Criterion that maximizes long-term log growth while constraining Prob(Minimum Wealth below alpha) below beta. In the example, basic Kelly position sizes ranged from 0 to 0.6, while risk-constrained Kelly ranged from 0 to 0.25, with lower volatility and drawdown but lower cumulative return.
Cold read: position sizing의 목적은 최대 수익이 아니라 game survival이다. 낮은 equity curve는 실패가 아니라 drawdown budget을 산 대가일 수 있다.
Trading insight: Henry가 conviction이 높을 때도 bet size는 edge estimate가 아니라 allowed drawdown probability로 cap을 씌워야 한다. full Kelly는 FOMO와 결합하면 ruin path가 된다.
Change sensor: rolling hit rate, win/loss payoff ratio, max drawdown probability, strategy heat.
Follow-up question: Henry 포트폴리오의 BTC, MSTR, BMNR에 fractional Kelly와 drawdown-constrained Kelly를 각각 적용하면 position cap이 어떻게 달라지는가?
[01:03 KST] Private Credit’s Software Lending Meets AI Disruption
Why it matters: AI는 equity upside narrative이면서 동시에 private credit collateral risk다. SaaS loan exposure가 private credit portfolios 안에 쌓이면, AI disruption은 public software stocks뿐 아니라 illiquid credit mark에도 영향을 준다. Henry가 AI theme을 단순 growth로만 보지 않게 만드는 credit-cycle sensor다.
Raw excerpt / abstract: BIS says outstanding loans to SaaS firms increased from almost USD 8 billion in 2015 to over USD 500 billion, or 19 percent of total direct loans, by end-2025. A third of private credit funds have extended loans to SaaS. Software stocks fell almost 30 percent between October 2025 and February 2026, while BDC stock prices fell about 10 percent and discounts to NAV deepened.
Cold read: AI shock는 생산성 shock인 동시에 debt underwriting shock다. illiquid private loans는 public equity보다 늦게 mark될 수 있어 hidden correlation이 생긴다.
Trading insight: AI 관련 risk-on bet은 semiconductor earnings만 보면 부족하다. BDC discount to NAV, private credit redemption pressure, SaaS default risk가 broader liquidity를 압박할 수 있다.
Why it matters: AI의 병목은 model idea만이 아니라 power, grid, transformer, cooling, construction capacity다. 이는 semiconductor demand와 utility, private credit, regional geopolitics를 하나로 묶는다. Henry가 AI narrative를 가격 momentum이 아니라 physical capacity race로 읽게 한다.
Raw excerpt / abstract: SemiAnalysis estimates global Datacenter Critical IT Power demand rises from 49 GW in 2023 to 96 GW by 2026, with AI consuming around 40 GW. It argues datacenter power capacity growth may accelerate from 12 to 15 percent CAGR to 25 percent CAGR, but the buildout is not smooth and a real power crunch is coming. It also estimates AI datacenters may use 4.5 percent of global energy generation by 2030.
Cold read: AI scale law is colliding with grid law. compute demand can be exponential in narrative, but power delivery and construction are lumpy, local, financed, and slow.
Trading insight: AI exposure should be stress-tested against power availability, grid connection timelines, capex financing, and utilization risk. The bottleneck can move from chips to megawatts to debt service.
Change sensor: announced MW versus energized MW, utility interconnection backlog, data center debt issuance, GPU utilization economics.
Follow-up question: AI infrastructure winners are better identified by chip shipment growth or by secured low-cost power and grid connection speed?
Run Summary [01:03 KST]
collected_count: 9
skipped_duplicates: 0
domain_mix: microstructure/execution 3, risk/cognition 2, macro/capital-markets 4, crypto-structure 2, AI-history 2. Several items intentionally overlap because execution quality, funding plumbing, and AI credit risk are converging.
top_theme: usable liquidity is not visible liquidity. Fill quality, ETF settlement, Treasury depth, repo plumbing, private credit marks, and datacenter power all show that headline capacity overstates executable capacity.
sharpest_insight: the same mistake repeats across domains: traders mistake reported flow, displayed depth, NAV, power pipeline, or Kelly fraction for usable capacity under stress.
danger_of_misuse: 오늘 자료를 bullish or bearish signal로 바로 바꾸면 안 된다. 이 자료들은 entry direction이 아니라 sizing cap, execution checklist, and regime filter로 써야 한다.
next_probe: Henry portfolio에 대해 adverse fill, 1 percent depth, Treasury liquidity stress, and drawdown-constrained Kelly를 하나의 pre-trade capacity score로 묶을 수 있는가?
[02:04 KST] The Cross-Border Trail of the Treasury Basis Trade
Why it matters: Treasury basis trade는 risk-free arbitrage처럼 보이지만 repo leverage, FICC sponsored repo, data undercount가 얽힌 shadow balance sheet다. Fed note는 Cayman hedge fund Treasury holdings가 TIC data에서 약 USD 1.4 trillion undercount될 수 있다고 말한다. Henry에게 중요한 점은 macro liquidity가 official flow table보다 훨씬 더 levered and fragile할 수 있다는 것이다.
Raw excerpt / abstract: FEDS Notes says Cayman-domiciled hedge funds purchased net USD 1.2 trillion of Treasury securities from January 2022 to December 2024 and absorbed 37 percent of net issuance of notes and bonds. It also says adding estimated undercount of roughly USD 1.4 trillion would make the Cayman Islands the largest foreign holder of U.S. Treasuries. The note links the measurement gap to repo collateral transfers and sponsored DVP repo.
Cold read: marginal Treasury buyer may not be household saving or stable foreign demand, but leveraged hedge fund basis balance sheet funded in repo. A smooth Treasury market can hide a crowded, collateral-sensitive trade until funding cost or volatility forces unwind.
Trading insight: BTC, MSTR, and high beta exposure should be throttled when Treasury basis leverage, sponsored repo volume, or repo funding stress rises. A Treasury liquidity shock can force cross-asset deleveraging even if crypto-specific thesis is unchanged.
Follow-up question: Can Henry’s risk dashboard infer a Treasury basis unwind warning from public CFTC, DTCC, SOFR, and Treasury liquidity data before risk assets react?
Why it matters: Crypto futures exchange risk is not only price direction; it is liquidation waterfall design. This paper frames ADL as optimization and shows why the most levered accounts should be reduced first under a water-filling rule. Henry should treat high leverage venues as endogenous risk systems where profitable accounts can still be forcibly delevered when the exchange loss buffer is insufficient.
Raw excerpt / abstract: The abstract states that ADL mechanisms reduce positions and socialize losses among solvent participants when margin and loss-absorbing resources are insufficient. In a single-asset isolated-margin setting, the unique optimal risk-neutral policy minimizes maximum leverage among participants, reducing positions first for the most highly levered accounts. In cross-margin multi-asset portfolios, naive gross leverage can be misleading because it ignores hedging.
Cold read: exchange design can turn other traders’ leverage into Henry’s execution risk. During stress, venue rules become part of the trade, not background plumbing.
Trading insight: Perp exposure should be sized not only by stop distance but by exchange ADL queue, insurance fund, cross-margin correlation, and gross versus hedged leverage. If venue ADL risk rises, position size or venue concentration should fall before price confirms stress.
Change sensor: insurance fund drawdown, liquidation volume, ADL queue indicators, cross-asset correlation spikes, funding plus OI concentration.
Follow-up question: Which venues used by Henry provide enough public ADL and insurance-fund data to justify meaningful leverage during volatility clusters?
[02:04 KST] Has the Liquidity Landscape for Bitcoin ETFs Changed?
Why it matters: ETF flow headlines can create false certainty because they aggregate long-term allocators, hedge funds, momentum funds, and arbitrage capital into one number. The article argues that the ETF channel has become a volatility transmission mechanism rather than a simple bullish adoption rail. Henry should avoid reading IBIT outflows or inflows as a single institutional belief signal.
Raw excerpt / abstract: Gate describes a USD 2.97 billion net outflow streak across spot BTC ETFs and notes that IBIT had a rare large outflow. It argues ETF channels now host capital with different holding behaviors and can transmit macro risk through ETF redemption, spot selling, derivatives liquidation, and further redemption feedback.
Cold read: ETF adoption did not remove BTC reflexivity; it moved reflexivity into a more visible TradFi wrapper. Flow transparency can amplify narrative speed in both directions.
Trading insight: ETF flow should be decomposed into persistence, issuer concentration, AP settlement lag, and on-chain custody movement before it changes position size. A one-day flow shock is not the same as a regime shift.
Change sensor: IBIT flow persistence, flow share by issuer, ETF turnover versus next-day flow, custody wallet movement to exchanges, derivatives liquidation cascade.
Follow-up question: What threshold separates normal ETF tactical rotation from true long-term allocator exit in BTC market structure?
Why it matters: DTCC sponsored repo volume is a public window into the same repo channel that Fed researchers link to Treasury basis trade undercount. This is not a directional trade signal; it is a leverage capacity gauge. Henry can use it as a plumbing sensor before assuming Treasury demand is stable.
Raw excerpt / abstract: DTCC says the charts show the amounts of repo and reverse repo activity of Sponsored Members on FICC’s platform, from the Sponsored Member’s perspective. The page provides downloadable SponsoredVolume.csv and historic data prior to April 1, 2020.
Cold read: if sponsored repo becomes the hidden balance sheet behind basis demand, the public CSV becomes more important than many macro opinion pieces. Plumbing data is not exciting, but it shows where leverage is actually housed.
Trading insight: Add sponsored repo volume trend and sudden changes to macro risk throttle. Rising volume with worsening Treasury liquidity or higher repo rates means more convex unwind risk.
Why it matters: Stablecoin liquidity is not just crypto cash; it is a reserve-asset and run-risk structure connected to bank deposits, Treasury repo, and AML compliance. Brookings highlights that weak capital, liquidity, reserve, and foreign-issuer rules could make stablecoins information-sensitive. Henry should not treat stablecoin market cap as always-available dry powder.
Raw excerpt / abstract: Brookings says U.S. dollar-backed stablecoins reached more than USD 260 billion in Q3 2025 and monthly transactions rose to more than USD 1 trillion. It warns that GENIUS allows uninsured bank deposits and credit union shares as reserves, creating two-way interconnections between bank risk and stablecoin risk. It also notes Tether had about 20 percent of reserve assets outside cash and cash equivalents in a cited statement.
Cold read: the crypto cash layer can become a money-market-fund style fragility layer if reserves are risky or illiquid. Stablecoin growth is liquidity only until redemption confidence becomes the trade.
Trading insight: BTC and alt liquidity assumptions need stablecoin reserve quality haircuts. During stress, stablecoin depeg or redemption friction can turn displayed exchange depth into unusable liquidity.
Change sensor: reserve composition attestations, issuer concentration, stablecoin market cap change, depeg frequency, bank deposit exposure.
Follow-up question: How should Henry haircut crypto venue depth when USDT or USDC reserve-risk headlines appear?
[02:04 KST] Unstable Coins: Stablecoin Regulation, Market Structure Legislation, and U.S. Security Risks
Why it matters: Crypto market structure is becoming geopolitical infrastructure, not just an asset class. CSIS argues that stablecoin and digital asset legislation can either preserve U.S. financial dominance or embed systemic and national security vulnerabilities. Henry should watch regulatory clarity as a capital-flow regime variable, not as a simple pro-crypto or anti-crypto headline.
Raw excerpt / abstract: CSIS says the GENIUS Act created the first comprehensive U.S. stablecoin framework but may create regulatory fragmentation, reserve gaps, and foreign issuer oversight risks. It also says market structure legislation may cement structural vulnerabilities, create regulatory arbitrage, and leave gaps in AML and sanctions enforcement for digital asset service providers and DeFi-related carveouts.
Cold read: regulatory clarity can attract capital, but bad clarity can institutionalize fragility. The key is not whether regulation passes; it is whether rule design reduces run risk, illicit finance risk, and arbitrage.
Trading insight: Crypto exposure should be sized around implementation details, not legislative slogans. The highest-risk period can be after a law passes but before capital, liquidity, reserve, and monitoring rules are proven.
Change sensor: foreign issuer safe harbor standards, AML coverage for service providers, stablecoin reserve requirements, DeFi carveout language.
Follow-up question: Which regulatory provisions would directly change BTC ETF demand, stablecoin liquidity, or U.S. exchange market depth?
[02:04 KST] How AI Debt Financing Impacts Duration Supply and Interest Rates
Why it matters: AI data center buildout is moving from equity narrative to fixed-income duration supply. Dallas Fed argues that AI infrastructure financing can steepen long-end curve dynamics through corporate bonds, pay-fixed swaps, and issuer substitution. Henry should see AI capex as a macro rates input, not only as NVDA-style equity demand.
Raw excerpt / abstract: Dallas Fed says data center investment estimates range between USD 3 trillion and USD 5 trillion over the next three to five years. Wall Street estimates of AI-related investment grade issuance center around USD 300 billion for 2026, potentially creating USD 360 billion in 10-year equivalents, about one eighth of Treasury duration supply. It also notes private credit floating-rate loans may be paired with pay-fixed swaps, creating synthetic duration supply.
Cold read: AI boom can pressure rates even if productivity optimism is real. Financing the buildout changes term premium, swap spreads, and credit markets before the revenue model is fully proven.
Trading insight: High beta tech and crypto trades should monitor long-end rates and swap spread divergence as AI financing sensors. AI upside narrative can coexist with a rates headwind that compresses multiple and raises funding cost.
Change sensor: AI-related IG issuance, 10s30s curve, 30-year swap spreads, private credit data center deals, hyperscaler capex guidance.
Follow-up question: Is AI capex financing becoming a hidden driver of term premium that could cap risk asset rallies even during strong earnings cycles?
[02:04 KST] What Hacker News Gets Right About AI Coding Agents in 2026
Why it matters: AI agents are shifting from model demos to workflow infrastructure, and the article’s key point is that verification capacity is the bottleneck. This maps directly to trading: idea generation is abundant, but review, risk control, and execution quality decide survival. Henry should avoid AI FOMO and focus on systems that make decisions more verifiable.
Raw excerpt / abstract: The article says the real product is workflow, not the model; skills are becoming more important than raw prompting; orchestration matters more than autonomy; and verification is the real bottleneck. It argues that teams should treat agents like workflow infrastructure, standardize project context, optimize for reviewability, and learn orchestration rather than just prompting.
Cold read: AI leverage is moving toward process design. The winners are not the loudest autonomy demos but the teams with repeatable workflows, bounded scopes, and deterministic checks.
Trading insight: Apply the same principle to trading research. More agents and more signals are only useful if every output goes through source check, duplication check, risk translation, and pre-trade checklist.
Change sensor: agent workflow adoption, verification tooling, repo-local skills conventions, production reliability metrics, pricing pressure on AI coding tools.
Follow-up question: What parts of Henry’s trading process can be safely agent-assisted without increasing false confidence or review debt?
Run Summary [02:04 KST]
collected_count: 8
skipped_duplicates: 1
domain_mix: microstructure/execution 3, risk/cognition 2, macro/capital-markets 5, crypto-structure 4, AI-history 2. Overlap is intentional because the strongest signals sit at the plumbing boundary between leverage, settlement, and financing.
top_theme: hidden leverage is migrating into official-looking wrappers: Treasury basis trades through sponsored repo, BTC through ETFs, stablecoins through regulated reserve rules, and AI through long-duration credit supply.
sharpest_insight: the safer an instrument looks from the outside, the more important its funding and settlement mechanics become; visible adoption can mask fragile leverage.
danger_of_misuse: 오늘 자료를 macro bearish로만 해석해 무조건 risk-off하면 안 된다. 핵심은 direction call이 아니라 leverage-sensitive sizing, venue selection, and execution timing이다.
next_probe: Treasury basis leverage, BTC ETF flow persistence, stablecoin reserve quality, and AI duration supply를 하나의 cross-asset hidden leverage dashboard로 묶을 수 있는가?
[03:02 KST] When AI Trading Agents Compete: Adverse Selection of Meta-Orders by Reinforcement Learning-Based Market Making
Why it matters: 이 논문은 AI agent가 시장에 들어오면 execution game이 단순히 더 빠른 주문으로 끝나지 않고 meta-order를 학습해 adverse selection을 만든다는 점을 보여준다. Henry가 큰 주문을 나누어 집행할 때, 상대방은 단순 market maker가 아니라 order flow를 학습하는 agent일 수 있다. execution quality는 점점 더 model versus model game이 된다.
Raw excerpt / abstract: The abstract says the authors use reinforcement learning within a Hawkes Limit Order Book model to replicate high-frequency market maker behavior. They test an RL agent against a medium-frequency trader executing a meta-order and show the RL market making agent learns to capitalise on price drift induced by the meta-order. It also notes that slippage costs for medium-frequency traders are likely to increase as high-frequency trading proliferates.
Cold read: 시장 impact는 주문 크기의 함수만이 아니라 상대 agent가 내 패턴을 얼마나 빨리 학습하는지의 함수가 된다. 나눠 사는 행동 자체가 signal이 될 수 있다.
Trading insight: Henry의 large entry는 fixed TWAP보다 randomized schedule, venue rotation, and post-fill drift measurement가 필요하다. execution edge가 없으면 좋은 thesis도 상대 agent에게 rent를 지불한다.
Why it matters: Treasury repo는 risk-free plumbing처럼 보이지만 haircut, margin, legal documentation, counterparty risk가 실제 resilience를 결정한다. NY Fed speech는 non-centrally cleared bilateral repo에서 zero haircut practices가 많았고 repo risk management recommendation completion target이 June 2026임을 언급한다. Henry의 high beta risk budget은 repo collateral system이 얼마나 안전하게 margining되는지와 연결된다.
Raw excerpt / abstract: The speech says the Treasury market is preparing for expanded central clearing, potential regulatory changes, and likely growth in Treasury supply. It notes NCCBR clearing and settlement arrangements are often bespoke and opaque, and OFR pilots found the majority of Treasury transactions in the segment had zero haircuts. It also says Treasury repo market daily average transaction volume is over USD 8 trillion.
Cold read: 가장 안전한 collateral의 funding market도 haircut competition과 opacity 때문에 hidden leverage를 쌓을 수 있다. resilience는 central clearing headline보다 margin discipline의 실제 adoption이다.
Trading insight: repo best practice rollout은 risk-on or risk-off signal이 아니라 leverage capacity reset이다. haircut adoption이 늘면 system resilience는 올라가지만 crowded basis trade의 economics는 바뀔 수 있다.
Henry Risk Link: narrative | liquidity | correlation
Why it matters: AI buildout is no longer only equity capex; it is becoming shadow borrowing through private placements, special purpose vehicles, leases, offtake agreements, and guarantees. BIS frames this as obligations economically like debt but often outside the hyperscaler balance sheet. Henry should read AI upside with credit transmission channels attached.
Raw excerpt / abstract: BIS says hyperscaler gross bond issuance topped USD 100 billion in 2025, mostly long-term. It describes off-balance sheet arrangements where a dedicated vehicle raises debt through private placements, while the hyperscaler holds a minority stake, commits to long-term operating leases or capacity offtake agreements, and may provide guarantees. BIS calls these arrangements shadow borrowing and warns they strengthen links between hyperscalers, private credit vehicles, insurers, and banks.
Cold read: AI capex can look asset-light at the public equity level while leverage migrates to vehicles and private credit. If project payoff uncertainty rises, the shock may appear first in CDS, private credit appetite, and guarantees rather than in model benchmarks.
Trading insight: AI-linked risk-on trades need a credit plumbing check: hyperscaler CDS, data center private credit spreads, lease guarantee disclosures, and bank funding lines. Narrative strength does not remove refinancing risk.
Change sensor: hyperscaler bond issuance, CDS spread divergence by rating, private credit data center deal terms, off-balance sheet guarantee disclosures.
Follow-up question: Which AI infrastructure exposures in public markets are actually leveraged credit trades disguised as secular growth equity?
[03:02 KST] Data Center Boom Expected to Raise Electricity Component of PCE Inflation
Why it matters: AI demand can become an inflation input through electricity prices, not only a productivity story. Dallas Fed estimates plausible data center buildout could add 0.04 to 0.13 percentage points to annual PCE inflation by 2030, while an extreme full-connection scenario could add 1.02 percentage points. This matters because higher inflation pressure can keep rates tighter and compress risk asset multiples.
Raw excerpt / abstract: Dallas Fed says power demand from data centers is projected by BloombergNEF to double from 40 GW today to 80 GW by 2031. Under plausible assumptions, annual PCE inflation in 2030 would increase by 0.04 to 0.13 percentage points through retail electricity prices. Slower renewables growth could nearly double the inflationary effect, and if all proposed data centers connected and operated fully, annual PCE inflation could rise 1.02 percentage points in 2030.
Cold read: AI is both deflationary software narrative and inflationary physical load. The market may price productivity while the grid prices scarcity.
Trading insight: AI-related positions should be stress-tested against electricity inflation, utility regulation, renewable buildout slippage, and rate sensitivity. A bullish AI earnings cycle can coexist with a macro rates headwind.
Change sensor: data center grid interconnection queue, wholesale electricity spikes, renewables project completion rate, state-level cost-shifting rules.
Follow-up question: At what point does AI load growth become a Fed-relevant inflation variable rather than a sector-specific utility issue?
[03:02 KST] Compute Accounts for the Majority of Expenses of AI Companies
Why it matters: Epoch AI shows compute is the dominant expense for AI companies, which makes AI businesses economically closer to capacity businesses than pure software. If R&D and inference compute are 54 percent to 62 percent of costs, margins depend heavily on utilization, pricing, hardware efficiency, and financing. Henry should not treat AI revenue growth as software-like gross margin by default.
Raw excerpt / abstract: Epoch AI estimates across Anthropic, Minimax, and Z.ai that R&D and inference compute together make up 54 percent to 62 percent of costs, while staff spending accounts for less than 25 percent. It says spending at each lab is 2 to 3 times higher than revenue. For Anthropic, it cites estimated USD 9.7 billion expenses, USD 4.1 billion training compute, and USD 2.7 billion inference compute expense for 2025.
Cold read: AI labs are scaling revenue and cost at the same time. The bottleneck is not just talent; it is compute economics and whether inference demand can cover capital intensity.
Trading insight: AI equity exposure should separate user growth from unit economics. Revenue acceleration without compute margin improvement may increase financing need rather than free cash flow.
Change sensor: inference gross margin, training spend versus revenue, token price cuts, model utilization, capex to revenue ratio.
Follow-up question: Which AI companies have pricing power enough to convert compute-heavy growth into durable free cash flow?
[03:02 KST] The Crypto Liquidity Concentration Report
Henry Risk Link: liquidity | execution | correlation
Why it matters: Kaiko shows crypto liquidity is concentrated across a small number of exchanges and concentrated differently for volume and depth. This matters because top-of-book availability can look healthy while exit capacity is dependent on a few venues. Henry should translate crypto liquidity into venue concentration risk, not just aggregate market depth.
Raw excerpt / abstract: Kaiko says the top 8 largest platforms accounted for 91.7 percent of global market depth and 89.5 percent of volume in its sample. It also says since the FTX collapse, 0.1 percent market depth recovered more than liquidity placed in wider ranges, while 1 percent, 2 percent, and 4 percent depth stayed mostly flat. Altcoin offshore depth concentration increased from 65 percent to 71 percent.
Cold read: crypto market maturity may mean liquidity is more professional, but also more centralized in failure points. Tight depth recovery is not the same as deep exit capacity.
Trading insight: BTC or alt sizing should haircut depth by venue concentration and distance from mid. For stress exits, 0.1 percent depth is less useful than 1 percent to 4 percent depth and venue reliability.
Change sensor: top 8 venue depth share, 0.1 percent versus 1 percent depth gap, offshore altcoin depth share, exchange-specific outage and withdrawal risk.
Follow-up question: Henry’s crypto exposure should be capped by aggregate liquidity or by the weakest reliable venue share during stress?
Why it matters: Stablecoin market cap is often read as crypto dry powder, but Coinbase’s framework highlights integration limits, Treasury demand effects, reserve rules, liquidity buffers, and redemption cascade risk. Stablecoin growth can deepen on-chain liquidity, but it can also create a money-market-like channel into T-bills and banking rails. Henry should treat stablecoin supply as conditional liquidity.
Raw excerpt / abstract: Coinbase says stablecoin integration into the existing financial system may limit total market size, including through net impact on U.S. Treasury demand. It says regulatory developments like the GENIUS Act are crucial for reserve rules and liquidity buffers, which could reduce the risk that large redemptions turn into a cascade of forced T-bill selling.
Cold read: stablecoins are becoming both payment rail and shadow cash instrument. Their liquidity value depends on reserve transparency, redemption mechanics, and whether regulation reduces or hides run risk.
Trading insight: Stablecoin supply growth should not automatically increase crypto risk budget. Apply reserve-quality haircut and watch redemption stress before treating it as deployable liquidity.
Why it matters: The key lesson is not that Kelly is optimal, but that unconstrained Kelly ignores path pain and survival. The paper turns drawdown probability into a constraint and creates a convex optimization approximation. Henry’s large bets should be gated by probability of breaching minimum wealth, not only expected return.
Raw excerpt / abstract: The abstract says the authors consider the classic Kelly gambling problem with general distribution of outcomes and an additional risk constraint that limits the probability of a drawdown of wealth to a given undesirable level. They develop a bound on drawdown probability; using this bound instead of the original risk constraint yields a convex optimization problem that guarantees the drawdown risk constraint holds.
Cold read: edge without drawdown constraint is not a strategy; it is a volatility tolerance test. The trader who cannot survive the path never realizes the long-run growth rate.
Trading insight: Position size should be the minimum of fractional Kelly, drawdown-constrained Kelly, liquidity cap, and correlation cap. This directly blocks FOMO-driven size escalation after wins.
Change sensor: drawdown probability target, minimum wealth threshold, rolling edge uncertainty, correlation-adjusted portfolio heat.
Follow-up question: What drawdown probability threshold should Henry accept for core BTC exposure versus tactical MSTR or BMNR exposure?
Run Summary [03:02 KST]
collected_count: 8
skipped_duplicates: 2
domain_mix: microstructure/execution 3, risk/cognition 2, macro/capital-markets 5, crypto-structure 2, AI-history 3. Overlap is intentional because the strongest overnight theme is that capacity becomes fragile when financed, concentrated, or learned by opposing agents.
top_theme: execution and infrastructure are becoming adversarial capacity problems. AI agents learn meta-orders, repo markets hide leverage in margin practices, stablecoins and crypto depth concentrate liquidity, and AI infrastructure converts narrative into power, duration, and shadow borrowing.
sharpest_insight: visible capacity is least reliable exactly when it becomes most valuable; order book depth, stablecoin supply, repo funding, and AI compute all need stress haircuts before they can support larger sizing.
danger_of_misuse: 오늘 자료를 AI bearish, crypto bullish, or macro bearish signal로 바로 바꾸면 안 된다. 올바른 사용법은 direction call이 아니라 sizing cap, liquidity haircut, execution randomization, and funding-stress filter다.
next_probe: Henry의 pre-trade checklist에 meta-order footprint, venue concentration haircut, repo haircut regime, stablecoin reserve score, and AI financing stress를 하나의 capacity-adjusted risk score로 통합할 수 있는가?