永续合约做市 (Hyperliquid / dYdX / GMX)

日期: 2026-07-27 方向: 量化 / 微观结构 / 做市 阶段: Phase 2 - 市场微观结构与做市 (Day 75-88) 标签: #Perp #Hyperliquid #dYdX #GMX #Funding #DEX

---

今日目标

永续合约 (perp) 是 crypto 的"主战场"——日均成交 $100B+，是现货 5-10 倍。三大 DEX perp 平台各有架构创新。

---

一、Perp 是什么？为什么做市机会丰富？

1.1 永续合约定义

• 没有到期日的期货合约
• 用 funding rate 机制让合约价格 ≈ spot price
• funding 每 8h（CEX）或 1h（Hyperliquid）/ 每 block（GMX）结算

1.2 funding rate 机制

premium = (perp_price - spot_price) / spot_price
funding_rate = clamp(premium + interest_rate, -0.75%, +0.75%)

every funding period:
  if funding_rate > 0:  long pays short
  else:                 short pays long

直觉：long > short → premium > 0 → long 付 funding 给 short → suppression 套利空间。

1.3 做市机会丰富的原因

1. Spread 普遍宽 vs 现货（liquidity 集中但 vol 大）
2. Funding rate 套利：spot long + perp short 锁差
3. Cross-venue mispricing：Binance vs Bybit vs DEX 不同时段差 5-50 bps
4. Liquidation cascading：清算时 spread 暴增 100x，做市可顺势 widen

---

二、三大 DEX Perp 架构对比

2.1 dYdX v4

• 全链上 LOB on Cosmos chain
• 撮合通过 validator network
• API 类似 CEX（限价挂单、IOC、stop）
• Maker 程序：volume share rewards

2.2 GMX v2

• LP 池（GLP / GM）做交易对手
• Trader 用 oracle 价格直接成交，无 spread 滑点（除 fee）
• LP 承担 OI imbalance 风险（funding 调节）
• 不是"传统"做市机会，但 LP 等于"做市"

2.3 Hyperliquid（重点）

• 独立 L1，自研共识
• 完全 onchain LOB（包括订单流）
• API 类似 Binance（REST + WS）
• 做市 incentive：HLP 池 + 季度 maker rewards

---

三、Hyperliquid 数据接入

3.1 REST API

Base URL: https://api.hyperliquid.xyz

POST /info
  body: { "type": <type>, ...filters }

types:
  - "meta"            -> universe, asset config
  - "l2Book"          -> orderbook snapshot
  - "allMids"         -> all asset mid prices
  - "metaAndAssetCtxs" -> per-asset funding, OI, mark, oracle
  - "candleSnapshot"  -> klines
  - "userState"       -> 自己的 position

3.2 真实样例响应：L2 Book

import requests
r = requests.post("https://api.hyperliquid.xyz/info",
                  json={"type": "l2Book", "coin": "BTC"}).json()
print(r)

{
  "coin": "BTC",
  "time": 1709337600000,
  "levels": [
    [
      {"px": "62150.0", "sz": "1.234", "n": 3},
      {"px": "62149.5", "sz": "0.500", "n": 1},
      {"px": "62149.0", "sz": "2.100", "n": 5}
    ],
    [
      {"px": "62150.5", "sz": "0.789", "n": 2},
      {"px": "62151.0", "sz": "1.456", "n": 4},
      {"px": "62151.5", "sz": "3.200", "n": 7}
    ]
  ]
}

levels[0] = bids, levels[1] = asks。每档 (price, size, num_orders)。

3.3 funding & open interest

r = requests.post("https://api.hyperliquid.xyz/info",
                  json={"type": "metaAndAssetCtxs"}).json()
# r[1] = list of asset contexts, in same order as r[0]['universe']
btc_idx = next(i for i,a in enumerate(r[0]["universe"]) if a["name"]=="BTC")
btc_ctx = r[1][btc_idx]

{
  "funding": "0.0000125",          // 1h funding rate
  "openInterest": "1234.567",     // BTC 单位
  "premium": "-0.000023",
  "oraclePx": "62148.50",
  "markPx": "62150.10",
  "midPx": "62150.25",
  "impactPxs": ["62149.0","62151.5"],  // 衡量大单 impact
  "dayNtlVlm": "234567890.12",
  "prevDayPx": "61234.50"
}

impactPxs 是 Hyperliquid 独有：估计 +-$10K size 立即 impact。

3.4 WebSocket

wss://api.hyperliquid.xyz/ws

subscribe message:
{
  "method": "subscribe",
  "subscription": {
    "type": "l2Book",
    "coin": "BTC"
  }
}

每次 push 完整 snapshot（不是增量）— 简化但 bandwidth 高
其它 subscription types:
  - "trades" coin
  - "candle" coin, interval
  - "bbo" coin (best bid/offer)
  - "userEvents" user (private)
  - "userFills" user

3.5 真实 trade event

{
  "channel": "trades",
  "data": [
    {"coin":"BTC","px":"62150.10","sz":"0.012",
     "side":"B","time":1709337600234,"hash":"0x..."}
  ]
}

side: "B" = buy aggressor, "A" = sell aggressor。

---

四、代码实现：hyper_data.py

"""
hyper_data.py — Hyperliquid 数据接入 + 做市机会分析
依赖：requests, pandas, numpy, websockets
"""
import requests, json, asyncio, websockets
import pandas as pd, numpy as np
from collections import deque

BASE_URL = "https://api.hyperliquid.xyz"
WS_URL = "wss://api.hyperliquid.xyz/ws"

# ----------------------------------------------------------
# 1. REST 函数
# ----------------------------------------------------------
def get_meta_ctxs():
    r = requests.post(f"{BASE_URL}/info",
                      json={"type":"metaAndAssetCtxs"}, timeout=5).json()
    universe = r[0]["universe"]
    ctxs = r[1]
    rows = []
    for a, c in zip(universe, ctxs):
        rows.append({
            "coin": a["name"],
            "max_lev": int(a.get("maxLeverage", 0)),
            "funding_1h": float(c.get("funding", 0)),
            "funding_8h_equiv": float(c.get("funding", 0)) * 8,
            "open_interest": float(c.get("openInterest", 0)),
            "premium": float(c.get("premium", 0)),
            "mark_px": float(c.get("markPx", 0)),
            "oracle_px": float(c.get("oraclePx", 0)),
            "mid_px": float(c.get("midPx", 0)),
            "day_vol_usd": float(c.get("dayNtlVlm", 0)),
            "prev_day_px": float(c.get("prevDayPx", 0)),
        })
    return pd.DataFrame(rows)

def get_l2_book(coin: str = "BTC", n_levels: int = 20):
    r = requests.post(f"{BASE_URL}/info",
                      json={"type":"l2Book","coin":coin}).json()
    bids = [(float(x["px"]), float(x["sz"]), x["n"]) for x in r["levels"][0][:n_levels]]
    asks = [(float(x["px"]), float(x["sz"]), x["n"]) for x in r["levels"][1][:n_levels]]
    return {
        "coin": coin,
        "time": r["time"],
        "bids": bids,
        "asks": asks,
        "best_bid": bids[0][0] if bids else None,
        "best_ask": asks[0][0] if asks else None,
        "spread_bps": (asks[0][0] - bids[0][0]) / asks[0][0] * 1e4 if (bids and asks) else None,
        "depth_5_bid": sum(s for _,s,_ in bids[:5]),
        "depth_5_ask": sum(s for _,s,_ in asks[:5]),
    }

def get_candles(coin="BTC", interval="1h", limit=24):
    r = requests.post(f"{BASE_URL}/info", json={
        "type":"candleSnapshot",
        "req":{"coin":coin,"interval":interval,
               "startTime": int(pd.Timestamp.now().timestamp()*1000)
                           - limit*3600*1000,
               "endTime": int(pd.Timestamp.now().timestamp()*1000)}
    }).json()
    df = pd.DataFrame(r)
    if df.empty: return df
    for c in ["o","h","l","c","v"]:
        df[c] = df[c].astype(float)
    df["t"] = pd.to_datetime(df["t"], unit="ms")
    return df

# ----------------------------------------------------------
# 2. 做市机会扫描
# ----------------------------------------------------------
def mm_opportunity_scanner(min_oi_usd=1e6, min_spread_bps=2.0):
    """
    扫所有 perp，找 spread 宽且有一定 OI 的做市候选
    """
    meta = get_meta_ctxs()
    out = []
    for _, row in meta.iterrows():
        oi_usd = row["open_interest"] * row["mark_px"]
        if oi_usd < min_oi_usd: continue
        try:
            book = get_l2_book(row["coin"])
        except Exception:
            continue
        if book["spread_bps"] is None or book["spread_bps"] < min_spread_bps:
            continue
        out.append({
            "coin": row["coin"],
            "spread_bps": book["spread_bps"],
            "oi_usd": oi_usd,
            "day_vol_usd": row["day_vol_usd"],
            "funding_1h": row["funding_1h"]*1e4,    # bps
            "depth_5_bid": book["depth_5_bid"],
            "depth_5_ask": book["depth_5_ask"],
        })
    return pd.DataFrame(out).sort_values("spread_bps", ascending=False)

# ----------------------------------------------------------
# 3. funding rate 套利监控
# ----------------------------------------------------------
def funding_arb_signal(coin="BTC", binance_funding_8h=None):
    """
    比较 Hyperliquid funding 与 Binance funding
    """
    meta = get_meta_ctxs()
    row = meta[meta["coin"] == coin].iloc[0]
    hl_funding_8h = row["funding_8h_equiv"] * 1e4   # bps
    if binance_funding_8h is None:
        # fetch Binance
        r = requests.get("https://fapi.binance.com/fapi/v1/premiumIndex",
                         params={"symbol":f"{coin}USDT"}).json()
        binance_funding_8h = float(r["lastFundingRate"]) * 1e4
    diff = hl_funding_8h - binance_funding_8h
    return {
        "hl_funding_8h_bps": hl_funding_8h,
        "binance_funding_8h_bps": binance_funding_8h,
        "diff_bps": diff,
        "annualized_arb_bps": diff * 3 * 365  # 8h 周期 365×3 倍
    }

# ----------------------------------------------------------
# 4. WebSocket 实时 spread 监控
# ----------------------------------------------------------
async def stream_bbo(coin="BTC", duration_s=60):
    spreads = deque(maxlen=1000)
    async with websockets.connect(WS_URL) as ws:
        await ws.send(json.dumps({"method":"subscribe",
                                  "subscription":{"type":"bbo","coin":coin}}))
        start = pd.Timestamp.now()
        while (pd.Timestamp.now() - start).total_seconds() < duration_s:
            raw = await ws.recv()
            msg = json.loads(raw)
            if msg.get("channel") == "bbo":
                bid = float(msg["data"]["bbo"][0]["px"])
                ask = float(msg["data"]["bbo"][1]["px"])
                spreads.append((ask - bid) / ((ask+bid)/2) * 1e4)
        return list(spreads)

# ----------------------------------------------------------
# 5. 演示
# ----------------------------------------------------------
if __name__ == "__main__":
    # 5.1 全市场扫描
    print("=== Hyperliquid Perp Markets ===")
    scan = mm_opportunity_scanner(min_oi_usd=1e6, min_spread_bps=1.0)
    print(scan.head(20).to_string())

    # 5.2 BTC L2 book
    print("\n=== BTC L2 Book ===")
    book = get_l2_book("BTC", n_levels=5)
    print(json.dumps({k:(v if not isinstance(v,float) else round(v,4))
                      for k,v in book.items() if k not in ["bids","asks"]},
                     indent=2))
    print("Top 5 bids:", book["bids"])
    print("Top 5 asks:", book["asks"])

    # 5.3 funding rate arb
    print("\n=== BTC Funding Arbitrage ===")
    arb = funding_arb_signal("BTC")
    print(arb)

    # 5.4 1h kline
    print("\n=== BTC 1h candles (last 24h) ===")
    df = get_candles("BTC", "1h", 24)
    print(df.tail(5))

预期输出：

=== Hyperliquid Perp Markets ===
   coin   spread_bps     oi_usd   day_vol_usd  funding_1h  depth_5_bid  depth_5_ask
0  HYPE   12.4           45000000  120000000   2.31         5000.0       4800.0
1  WIF    8.2            32000000  88000000    1.85         12000.0      11500.0
2  PURR   6.5            12000000  35000000    0.95         8000.0       7500.0
...

=== BTC L2 Book ===
{
  "coin":"BTC", "time": 1709337600000,
  "best_bid": 62150.0, "best_ask": 62150.5,
  "spread_bps": 0.0805, "depth_5_bid": 5.234, "depth_5_ask": 5.789
}
...

=== BTC Funding Arbitrage ===
{'hl_funding_8h_bps': 1.0, 'binance_funding_8h_bps': 0.5,
 'diff_bps': 0.5, 'annualized_arb_bps': 547.5}

4.1 多 venue 同步监控（生产 mm 必需）

async def cross_venue_monitor():
    """
    同时订阅 Hyperliquid + Binance + Bybit BTC，找跨所价差
    """
    venues = {
        "HL": ("wss://api.hyperliquid.xyz/ws",
               {"method":"subscribe",
                "subscription":{"type":"bbo","coin":"BTC"}}),
        "Binance": ("wss://fstream.binance.com/ws/btcusdt@bookTicker", None),
        "Bybit": ("wss://stream.bybit.com/v5/public/linear",
                  {"op":"subscribe","args":["orderbook.1.BTCUSDT"]}),
    }
    # 并发订阅，组合 mid 计算 cross-venue mispricing
    ...

---

五、做市策略：Hyperliquid 适应

5.1 主要差异 vs Binance

5.2 策略调整

1. Wider initial spread — 200ms latency 让 stale quote 风险高，比 Binance 多加 2-3 bps
2. Lower fill rate target — order count vs fill ratio 接受更低（因 latency）
3. Funding-aware：1h funding 让 funding-arb 信号更新更频繁
4. Cross-venue mid：用 Binance mid + premium 估 fair value，而非纯 HL mid

5.3 HLP (Hyperliquid Pool) 做市机会

HLP 是 HL 的"被动 LP" 池：

• 类 GMX 但做 LOB 后端
• 任何人可存 USDC 进 HLP，分润成交 PnL + funding
• HLP 在 LOB 里挂被动单（被吃即赚 spread）

特殊机会：HLP 的"吃单速率"由协议参数决定，可计算 expected APY。

---

六、CEX vs DEX Perp 对比 (深度)

6.1 哪类做市机会在哪里更好？

---

七、风险与陷阱

1. Hyperliquid sequencer 中心化 单点失败 → 一段时间不可交易，库存裸暴露。修复：HL 头寸不超过总 portfolio 的 30%。

2. Funding rate spikes meme coin perp 可能 funding 50 bps/8h（瞬态 spike）。多头一夜亏 5%。修复：funding > 阈值时强制减仓。

3. Oracle delay (GMX) GMX 用 Chainlink，1 min 才更新。GMX MM bot 在 oracle 跳前抢交易 = 显著 alpha，但被官方限制。

4. HL withdrawal 队列 USDC 出 HL 桥到 ETH 主网有 4 hr 延迟（fraud proof window）。临时缺资 → 不能 hedge → 库存裸。

5. API rate limit 不一 Binance 1200 weight/min；HL 比较宽松但 burst 限制不同。生产 MM 需要 venue-specific rate limit handler。

6. DEX 上 long tail 流动性 illusion 某 meme perp display $5M OI 但实际 50 个 wallet。一个 whale 出走 spread 立刻 100x。

7. 抽税 / 提现限制 美国 / 欧洲监管影响 dYdX、GMX 用户。HL 仍 grey area。生产策略要考虑 KYC tier。

---

八、关键速查

Hyperliquid endpoints:
  POST https://api.hyperliquid.xyz/info     (read)
  POST https://api.hyperliquid.xyz/exchange (write, signed)
  WS   wss://api.hyperliquid.xyz/ws

Info types:
  l2Book / allMids / metaAndAssetCtxs / candleSnapshot
  userState / userFills / userFunding

WS subscriptions:
  l2Book / trades / candle / bbo / userEvents

Funding：
  perp = spot · (1 + premium)
  premium > 0 → long pays short

Cross-venue arb expected:
  diff_bps · 3 funding_per_day · 365 = APY bps

三大 DEX perp 速记

dYdX v4   — 完全 LOB on Cosmos chain，CEX-like API
GMX v2    — LP 池被动接，oracle priced，无 spread
Hyperliquid — 全链上 LOB，自研 L1，最接近 Binance

---

九、面试题

Q1: Hyperliquid 与 Binance Futures 的核心架构差异？

A: (i) HL 是独立 L1，所有订单流上链；Binance 集中式 matching engine。(ii) HL 自研共识，200ms block time；Binance µs；(iii) HL maker/taker fee 类似，但 maker rebate 程序通过 HLP 分润，间接给做市商；(iv) HL withdrawal 经 ETH 桥需 fraud-proof window；(v) HL 完全无 KYC，Binance 视司法管辖区严格。

Q2: GMX LP（GLP）如何"做市"？

A: GLP 持有多种资产（ETH、BTC、USDC等），按 weights。Trader 用 oracle 价格直接和 GLP swap/open position。Trader 长期 PnL → GLP 短期承担（trader 赢 GLP 亏，反之）。"做市"在这里 = 提供 capital 让 trader 有对手。LP 收益 = trading fee + funding 不平衡的 net + price exposure to GLP basket。

Q3: 跨 venue funding arb 如何执行？

A: 例：HL BTC funding 1 bps/8h、Binance 0.5 bps/8h，差 0.5 bps。 Strategy: Binance long BTC perp + HL short BTC perp，每 8h 净收 0.5 bps。 Steps: (i) 对 USDT/USDC 各放 50% 抵押；(ii) 锁仓 size 相同；(iii) 监控两边的 funding，任一出现负值时关仓；(iv) 总 PnL ≈ funding diff − fee × 2 − spread × 2。年化 0.5 bps × 3 × 365 = 547 bps（5.5% APY），高位 carry。

Q4: 在 HL 做市策略与 Binance 做市策略的关键不同点？

A: (i) latency：HL 200ms vs Binance < 10ms → HL 用 wider spread；(ii) funding 周期 1h vs 8h → HL funding 信号更高频；(iii) cancellation 限制：HL 鼓励 batch；(iv) HL 用户更窄群体（crypto-native），noise 流量比 Binance 小 → adverse selection 比例高 → 必须更保守 inventory；(v) HL 资产范围有限（150+ vs Binance 300+），策略 cross-asset 选择少。

Q5: 评估一个 DEX perp 平台的"做市友好度"，你会看什么？

A: 检查项：

1. maker fee schedule — rebate vs flat
2. API quality — REST + WS 完整性、文档、stability
3. rate limit — 能否支持 high-freq cancel/replace
4. funding rate machanism — 是否易被 manipulation
5. liquidation engine fairness — 是否给 maker 优先 fill
6. deposit/withdraw friction — 资金能否 30min 内进出
7. maker rewards program — 季度/月度 maker volume rewards
8. OI cap & tier — 大户进入会否被 reject
9. DEX 是否被 sequencer 抢跑 — privacy guarantees
10. historical uptime + incident response

---

明日预告

Day 88：Week 13 复习与 DEX vs CEX 做市完整对比 — Phase 2 收官。把 14 天的内容压缩成一份**"DEX vs CEX 做市对比报告"**：架构、数学、策略、收益、风险五维度。这份报告将作为简历/面试用的 capstone 产出，证明你掌握了从 LOB 微观结构到 V3 LP 数学到永续 hedge 的完整链条。