The Highest Funding Rate on Binance Is Usually a Delisting in Disguise

Funding-rate arbitrage is supposed to be delta-neutral. You hedge price, so price stops being the thing that hurts you. What hurts you instead is the venue changing the rules underneath the position: the perpetual contract gets settled, the spot pair gets removed, margin borrowing gets disabled, deposits stop being credited, withdrawals close. Every one of those events is announced in advance, in public, on a URL anyone can poll. Almost nobody polls it. So I open-sourced the boring part — a Binance announcement scraper that filters for delisting notices, extracts the affected tickers, deduplicates against a state file, and pushes to a webhook: github.com/godzilla-foundation/godzilla-community → strategies/binance_delisting_monitor One file, one dependency (requests). This post is about why it exists. The carry trade, and where it actually breaks The basic funding-rate carry is unglamorous on purpose. Long the spot asset, short the perpetual contract on the same asset, size the legs so net delta is roughly zero, and collect the funding payment every settlement interval. Price goes wherever it wants; you get paid for warehousing the imbalance between leveraged longs and the spot market. Because the price risk is hedged, the residual risk set is entirely operational: Basis risk — the two legs don't converge when you want them to. Execution risk — you leg in badly, or you can't leg out at all because the book is gone. Margin risk — a liquidation on the perp leg turns a hedged position into a directional one. Venue lifecycle risk — the instrument itself stops existing. The first three get a lot of attention because they show up in backtests. The fourth doesn't show up in backtests at all, because most historical datasets quietly drop symbols that no longer exist. That's textbook survivorship bias, and in perp markets it's severe: the universe of listed contracts turns over constantly. Your beautiful funding-yield backtest was run on the survivors. Why the top of the leaderboard is adversely selected Here's the uncomfortable part. Rank every USDⓈ-M perpetual by absolute funding rate, annualize it, and look at the top ten. On any given day a meaningful slice of that list is made up of contracts that are thin, distressed, or already carrying a warning label. That is not a coincidence — it's the mechanism working exactly as designed. Extreme funding is the price of an extreme positioning imbalance, and positioning goes lopsided precisely when a token is in trouble. A name that has been flagged for review attracts a wall of shorts. Perp trades at a persistent discount to spot. Funding goes deeply negative. Annualized, that prints a number in the hundreds of percent, and it lands at the top of every screener on earth. Then you have to actually harvest it, and this is where the shape of the trade matters: Funding positive (longs pay shorts): harvest with long spot + short perp. Your short leg is the contract that is about to be settled out of existence. Funding negative (shorts pay longs): harvest with long perp + short spot. Your short-spot leg requires borrowing the token on margin — and disabling margin borrow is one of the first things the exchange does on the way to a delisting, typically a day or two before removal. Either direction, the leg you depend on is the leg the exchange is actively dismantling. You are being paid a spectacular rate to hold an instrument whose remaining life is measured in days, and the payment stream terminates on a schedule that was published before you entered. Four ways a delisting kills a hedged position 1. The hedge dies before the spot leg does This is the one that actually blows people up, and it's a pure calendar problem. Delisting waves don't remove everything at once — the futures contracts are settled first, spot trading stops later, and withdrawals close later still. In one 2026 wave, futures positions were settled on March 24 while spot trading didn't stop until April 1 and the withdrawal window ran to June 1. Read that as a cash-and-carry operator. On day one your short perp is force-settled. From that moment you are not running a delta-neutral carry — you are holding a naked long in a token that the largest venue in the world just publicly condemned, and you have to sit in it while everyone else is selling into the same eight-day exit window. The hedge is gone; the exposure is not. 2. Auto-settlement happens at the exchange's mark, not at your price Binance's futures delisting notices follow a consistent template: no new positions from a cutoff time (commonly 30 minutes before), then all open positions closed and automatically settled at the delisting timestamp. If you didn't unwind, you get the mark price at that instant. For an illiquid contract in the middle of a delisting-driven repricing, the mark at a single arbitrary second is not a price you would have chosen. Your unwind slippage stops being a function of your execution quality and becomes a function of what the book looked like at 09:00 UTC. 3. The funding number itself can be changed administratively This is the detail that should genuinely bother anyone extrapolating a funding curve. The standard futures delisting announcement reserves the right to take additional protective measures on the affected contracts without further announcement — including adjusting leverage, position limits, and maintenance margin, changing funding-rate parameters such as the interest rate component, the premium component and the capped funding rate, altering the constituents of the price index, and switching the mark price to a protected last-price mechanism. Every input to your expected-return calculation is on that list. The funding cap can move. The index the premium is computed against can be recomposed. The mark price methodology can change. You are not modeling a market variable; you are modeling an administrative decision, and the announcement told you in advance that the decision could arrive silently. 4. Collateral, deposits and withdrawals — the plumbing risk The parts that hurt after the trade is over: Deposits stop being credited shortly after spot trading ends. If your rebalancer is mid-transfer, the asset can land somewhere that no longer accepts it. Withdrawals close on a separate, later date. Miss it and remaining balances may be converted to stablecoins — but the exchange is careful to say may, not will. That is not a fill you can plan around. Related products unwind on their own schedules: margin pairs, loans, Simple Earn positions, copy trading, Convert support. Each has its own cutoff. If you run cross-venue — long spot on Binance, short perp somewhere else — a Binance-side delisting can strand the leg you need to deliver while the other venue's contract keeps running. Now you're closing at Binance's terms on one side and at market on the other. None of this is exotic. It's all in the announcement. The only real failure is not reading it in time. The early-warning chain The useful thing about Binance specifically is that delistings are rarely a surprise if you're watching the right feed. There's a fairly consistent escalation: Monitoring Tag — the asset is flagged as higher-risk and under closer review. Tagging does not guarantee removal, but in the August 2026 wave, all six delisted tokens had carried the tag beforehand, with lead times ranging from roughly three and a half weeks to over fifteen. Vote to Delist — a public community mechanism that produces an explicit list of candidates, though the exchange has been clear that the vote is one input among several rather than a deterministic rule. Product-level suspensions — margin pairs, loans, and Earn support get pulled for a name before the token itself goes. The delisting announcement, with the full timetable. Futures settlement → spot halt → deposit cutoff → withdrawal close. For a systematic operator, stages 1 through 3 are the actual signal. By stage 4 you're just executing a fire drill. What the monitor does The script is deliberately small. It polls Binance's public CMS announcement endpoints, walks the announcement catalogs, regex-filters titles for delisting language, pulls out the affected tickers, and remembers what it has already reported. Endpoint fallbacks. The announcement API only lives on the www hosts, and there are two generations of the article endpoint with different response shapes. The client tries each in order and caches the first host that works: python API_HOSTS = ("https://www.binance.com", "https://www.binance.info") ARTICLE_PATHS = ( "/bapi/composite/v1/public/cms/article/list/query", "/bapi/apex/v1/public/apex/cms/article/list/query", CATALOG_PATH, ) It also handles the two different response envelopes — some versions return data.articles, others nest articles under data.catalogs[*].articles — and backs off on HTTP 429. Bilingual matching. Binance's Chinese announcements don't say "delist," they say 下架 / 停止交易. Both languages are covered, along with an exclusion list so that "resumption of trading" and 取消下架 notices don't fire a false alarm: python DELIST_PATTERNS = [ r"de-?list", r"\bremov(?:e|es|ed|al)\b", r"ceas(?:e|es|ing)\s+(?:the\s+)?trading", r"terminat\w*\s+(?:the\s+)?(?:trading|support|service)", r"suspend\w*\s+(?:the\s+)?(?:spot\s+)?trading", r"下架", r"停止.{0,4}交易", r"终止.{0,6}(?:交易|服务|支持)", ] EXCLUDE_PATTERNS = [ r"cancel\w*\s+the\s+(?:delisting|removal)", r"resum\w*\s+(?:the\s+)?trading", r"取消下架", r"恢复交易", ] Ticker extraction. Titles come in three shapes — full pairs (AGLDUSDT), tickers in parentheses (Across Protocol (ACX)), and bare comma-separated lists (Will Delist AGLD, LOOM, POLS, RARE). The bare case is the common one and also the noisiest, since announcement titles are Title Case and plenty of all-caps words aren't tickers. Hence a NOT_TICKERS set covering API, APR, KYC, SEPA, currency codes, country codes and so on. Idempotency. Announcement codes already reported are written to a JSON state file with an atomic replace, so a cron job that gets killed mid-write doesn't corrupt state and a daily run only ever emits what's new. That's the whole difference between an alert you act on and an alert you learn to ignore. Delivery. Text or JSON to stdout, JSONL append to a file, or a webhook push with the payload shape auto-detected from the URL (Feishu/Lark, DingTalk, WeCom, Slack/Discord). bash # daily cron 0 9 * * * cd /opt/monitors && python3 binance_delisting_monitor.py \ --webhook "$BINANCE_WEBHOOK_URL" >> monitor.log 2>&1 # machine-readable, for piping into a risk service python3 binance_delisting_monitor.py --json # connectivity check — the announcement domain gets TLS-reset on some networks python3 binance_delisting_monitor.py --selftest --proxy http://127.0.0.1:7897 Wiring it into an arb stack Alerting a human is the weak version. The strong version is a policy the strategy enforces without asking anyone: Hard symbol gate. Any ticker extracted from a delisting hit goes into a blacklist the strategy reads before opening a position. No new carry in a condemned name, regardless of what the funding rate says. This alone removes most of the tail. Unwind trigger, not an alert. If the blacklist intersects current positions, schedule an unwind against the earliest published deadline, which for a hedged position is the futures settlement, not the spot halt. Unwinding into the announcement window is bad. Unwinding at auto-settlement is worse. Rank penalty, not just exclusion. Feed the Monitoring Tag and Vote-to-Delist lists in as a score haircut on the funding-yield ranking. A contract with a 300% annualized rate and a warning label should not outrank a boring one at 25% that will still exist next month. The general principle: your expected funding yield is only realizable over the instrument's remaining lifetime. If you don't model that lifetime, you're not computing a yield — you're computing a rate and hoping. What this does not do In the interest of not overselling a 21KB script: Titles only. It doesn't fetch announcement bodies, so it won't extract exact settlement timestamps or the full affected-pair list when the title is generic. It tells you something is happening to X, and a human or a second stage still has to read the notice. Polling, not push. Daily by default. Fine for delistings, which come with days or weeks of notice. Useless for anything that needs sub-minute reaction. Binance only. No OKX, Bybit, Bitget, no exchange Twitter accounts, no on-chain signals. If you run multi-venue, this is one input. Unofficial endpoints. The CMS API is what the announcement page itself calls. It is not a documented, versioned API, and it can change without notice. The multi-path fallback exists because it already has. False positives are real. Keyword matching on titles will occasionally flag a routine notice. That is the correct bias for a risk monitor — but if you auto-blacklist on hits, expect to occasionally exclude something you didn't need to. The closing argument Delta-neutral doesn't mean risk-free. It means you've traded a risk you can see on a chart for a set of risks that only appear in announcement pages, margin schedules and settlement calendars. The funding rate leaderboard is a genuinely useful tool. It's just that, on any given day, part of what it's actually ranking is how close each contract is to not existing. Reading the announcements is the cheapest correction available — a cron job and a webhook, against a failure mode that turns a hedged book into a naked long overnight. The code is here. Issues and PRs welcome, especially for other venues. Disclaimer: Nothing here is investment advice. Exchange policies, endpoints, and timelines change; verify against the official announcement before acting on anything in this post.