Add Avellaneda-Stoikov market making strategy example

examples/avellaneda_stoikov_strategy.py

@@ -0,0 +1,396 @@
+"""
+Avellaneda-Stoikov Market Making Strategy
+
+Optimal market making with inventory management based on:
+"High-frequency trading in a limit order book" - Avellaneda & Stoikov (2008)
+Reference: https://github.com/fedecaccia/avellaneda-stoikov
+
+Computes a reservation price (inventory-adjusted fair value) and optimal
+spread to balance profit capture against inventory risk. The reservation
+price shifts away from the mid-price proportionally to inventory, encouraging
+mean-reversion of the position toward zero.
+
+Key formulas:
+    Reservation price: r = s - q * gamma * sigma^2 * (T - t)
+    Optimal spread:    spread = (2 / gamma) * ln(1 + gamma / k)
+    Ask quote:         r + spread / 2
+    Bid quote:         r - spread / 2
+
+Where:
+    s     = mid-market price
+    q     = current inventory (positive = long)
+    gamma = risk aversion (higher = tighter inventory control)
+    k     = order arrival sensitivity (higher = tighter spread)
+    sigma = price volatility over the trading session
+    T - t = remaining time fraction [0, 1]
+"""
+
+import argparse
+import math
+import os
+import sys
+import time
+from collections import deque
+from typing import Dict, List, Optional
+
+from dotenv import load_dotenv
+
+from dr_manhattan import Strategy
+from dr_manhattan.base import Exchange, create_exchange
+from dr_manhattan.models import Market
+from dr_manhattan.models.order import OrderSide
+from dr_manhattan.utils import prompt_market_selection, setup_logger
+from dr_manhattan.utils.logger import Colors
+
+logger = setup_logger(__name__)
+
+
+class AvellanedaStoikovStrategy(Strategy):
+    """
+    Avellaneda-Stoikov optimal market making strategy.
+
+    Quotes bid and ask prices derived from a reservation price that accounts
+    for inventory risk, and an optimal spread that balances profit against
+    adverse selection.
+
+    The reservation price shifts away from the mid-price as inventory grows,
+    encouraging mean-reversion of the position toward zero. As time remaining
+    decreases, the inventory penalty shrinks, allowing tighter quotes near
+    the end of the session.
+
+    Parameters:
+        gamma: Risk aversion. Higher values penalize inventory more heavily,
+               producing more aggressive mean-reversion but potentially
+               missing trades. Range: 0.1 - 1.0 for prediction markets.
+        k:     Order arrival sensitivity. Controls how much the spread
+               widens beyond the minimum. Higher values produce tighter
+               spreads. Range: 10 - 100 for prediction markets.
+        time_horizon_hours: Trading session length. The inventory penalty
+               decays to zero as the session ends.
+        volatility_window: Number of ticks used to estimate price volatility.
+    """
+
+    def __init__(
+        self,
+        exchange,
+        market_id: str,
+        gamma: float = 0.5,
+        k: float = 50.0,
+        time_horizon_hours: float = 1.0,
+        volatility_window: int = 30,
+        min_spread: float = 0.0,
+        max_position: float = 100.0,
+        order_size: float = 5.0,
+        max_delta: float = 20.0,
+        check_interval: float = 5.0,
+    ):
+        super().__init__(
+            exchange=exchange,
+            market_id=market_id,
+            max_position=max_position,
+            order_size=order_size,
+            max_delta=max_delta,
+            check_interval=check_interval,
+            track_fills=True,
+        )
+
+        self.gamma = gamma
+        self.k = k
+        self.time_horizon = time_horizon_hours * 3600.0
+        self.volatility_window = volatility_window
+        self.min_spread = min_spread
+
+        self.price_history: Dict[str, deque] = {}
+        self.start_time: float = 0.0
+
+    def setup(self) -> bool:
+        if not super().setup():
+            return False
+
+        self.start_time = time.time()
+
+        for outcome in self.outcomes:
+            self.price_history[outcome] = deque(maxlen=self.volatility_window)
+
+        spread = self._optimal_spread()
+
+        logger.info(
+            f"\n{Colors.bold('Avellaneda-Stoikov Config:')}\n"
+            f"  gamma: {Colors.yellow(f'{self.gamma:.3f}')} | "
+            f"k: {Colors.yellow(f'{self.k:.1f}')} | "
+            f"T: {Colors.cyan(f'{self.time_horizon / 3600:.1f}h')}\n"
+            f"  Base spread: {Colors.green(f'{spread:.4f}')} | "
+            f"Min spread: {Colors.green(f'{self.min_spread:.4f}')} | "
+            f"Vol window: {Colors.gray(f'{self.volatility_window} ticks')}"
+        )
+
+        return True
+
+    def on_tick(self) -> None:
+        self.refresh_state()
+
+        for ot in self.outcome_tokens:
+            outcome = ot.outcome
+            token_id = ot.token_id
+
+            bid, ask = self.get_best_bid_ask(token_id)
+            if bid is None or ask is None or bid <= 0 or ask <= 0:
+                continue
+
+            mid_price = (bid + ask) / 2.0
+            self.price_history[outcome].append(mid_price)
+
+            sigma = self._estimate_volatility(outcome)
+            if sigma is None:
+                continue
+
+            elapsed = time.time() - self.start_time
+            remaining = max(0.0, 1.0 - elapsed / self.time_horizon)
+
+            q = self._positions.get(outcome, 0.0)
+
+            # Reservation price: mid adjusted for inventory risk
+            reservation = mid_price - q * self.gamma * sigma**2 * remaining
+
+            # Optimal spread with floor
+            spread = max(self._optimal_spread(), self.min_spread)
+
+            ask_price = reservation + spread / 2.0
+            bid_price = reservation - spread / 2.0
+
+            # Clamp to valid prediction market range
+            ask_price = self.round_price(max(self.tick_size, min(1.0 - self.tick_size, ask_price)))
+            bid_price = self.round_price(max(self.tick_size, min(1.0 - self.tick_size, bid_price)))
+
+            if bid_price >= ask_price:
+                continue
+
+            self._update_quotes(outcome, token_id, bid_price, ask_price, q)
+
+        self.log_status()
+
+    def _estimate_volatility(self, outcome: str) -> Optional[float]:
+        """
+        Estimate price volatility from recent absolute price changes,
+        scaled to the full trading session.
+        """
+        history = self.price_history.get(outcome)
+        if not history or len(history) < 2:
+            return None
+
+        prices = list(history)
+        changes = [prices[i] - prices[i - 1] for i in range(1, len(prices))]
+
+        if not changes:
+            return None
+
+        mean = sum(changes) / len(changes)
+        variance = sum((c - mean) ** 2 for c in changes) / len(changes)
+        sigma_per_tick = math.sqrt(variance) if variance > 0 else 0.0
+
+        # Scale tick volatility to full-period volatility
+        ticks_per_period = self.time_horizon / self.check_interval
+        sigma = sigma_per_tick * math.sqrt(ticks_per_period)
+
+        return max(sigma, 0.001)
+
+    def _optimal_spread(self) -> float:
+        """Compute optimal spread: (2 / gamma) * ln(1 + gamma / k)"""
+        return (2.0 / self.gamma) * math.log(1.0 + self.gamma / self.k)
+
+    def _update_quotes(
+        self,
+        outcome: str,
+        token_id: str,
+        bid_price: float,
+        ask_price: float,
+        position: float,
+    ) -> None:
+        """Cancel stale orders and place new bid/ask at computed prices."""
+        buy_orders, sell_orders = self.get_orders_for_outcome(outcome)
+
+        self.cancel_stale_orders(buy_orders, bid_price)
+        self.cancel_stale_orders(sell_orders, ask_price)
+
+        if not self.has_order_at_price(buy_orders, bid_price):
+            if position + self.order_size <= self.max_position and self.cash >= self.order_size:
+                try:
+                    self.create_order(outcome, OrderSide.BUY, bid_price, self.order_size, token_id)
+                    self.log_order(OrderSide.BUY, self.order_size, outcome, bid_price)
+                except Exception as e:
+                    logger.error(f"    BUY failed: {e}")
+
+        if not self.has_order_at_price(sell_orders, ask_price):
+            if position >= self.order_size:
+                try:
+                    self.create_order(outcome, OrderSide.SELL, ask_price, self.order_size, token_id)
+                    self.log_order(OrderSide.SELL, self.order_size, outcome, ask_price)
+                except Exception as e:
+                    logger.error(f"    SELL failed: {e}")
+
+
+def find_market_id(
+    exchange: Exchange,
+    slug: str,
+    market_index: Optional[int] = None,
+) -> Optional[str]:
+    """Find market ID by slug/keyword search with optional selection."""
+    logger.info(f"Searching for market: {slug}")
+
+    markets: List[Market] = []
+
+    if hasattr(exchange, "fetch_markets_by_slug"):
+        markets = exchange.fetch_markets_by_slug(slug)
+
+    if not markets:
+        keywords = slug.replace("-", " ").lower()
+        keyword_parts = [k for k in keywords.split() if len(k) > 2]
+
+        all_markets: List[Market] = []
+        for page in range(1, 6):
+            try:
+                page_markets = exchange.fetch_markets({"page": page, "limit": 20})
+                if not page_markets:
+                    break
+                all_markets.extend(page_markets)
+            except Exception:
+                break
+
+        markets = [m for m in all_markets if all(k in m.question.lower() for k in keyword_parts)]
+
+    if not markets:
+        logger.error(f"No markets found for: {slug}")
+        return None
+
+    if len(markets) == 1:
+        logger.info(f"Found: {markets[0].question}")
+        return markets[0].id
+
+    if market_index is not None:
+        if 0 <= market_index < len(markets):
+            logger.info(f"Selected [{market_index}]: {markets[market_index].question}")
+            return markets[market_index].id
+        else:
+            logger.error(f"Index {market_index} out of range (0-{len(markets) - 1})")
+            return None
+
+    return prompt_market_selection(markets)
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description="Avellaneda-Stoikov Market Making Strategy")
+    parser.add_argument(
+        "-e",
+        "--exchange",
+        default=os.getenv("EXCHANGE", "polymarket"),
+        help="Exchange name (default: polymarket)",
+    )
+    parser.add_argument("-m", "--market-id", default=os.getenv("MARKET_ID", ""), help="Market ID")
+    parser.add_argument("-s", "--slug", default=os.getenv("MARKET_SLUG", ""), help="Market slug")
+    parser.add_argument(
+        "--market", type=int, default=None, dest="market_index", help="Market index"
+    )
+    parser.add_argument(
+        "--gamma",
+        type=float,
+        default=0.5,
+        help="Risk aversion parameter (default: 0.5)",
+    )
+    parser.add_argument(
+        "--k",
+        type=float,
+        default=50.0,
+        help="Order arrival sensitivity (default: 50.0)",
+    )
+    parser.add_argument(
+        "--time-horizon",
+        type=float,
+        default=1.0,
+        help="Trading session length in hours (default: 1.0)",
+    )
+    parser.add_argument(
+        "--vol-window",
+        type=int,
+        default=30,
+        help="Volatility estimation window in ticks (default: 30)",
+    )
+    parser.add_argument(
+        "--min-spread",
+        type=float,
+        default=0.0,
+        help="Minimum spread floor (default: 0.0)",
+    )
+    parser.add_argument(
+        "--max-position",
+        type=float,
+        default=float(os.getenv("MAX_POSITION", "100")),
+        help="Maximum position size (default: 100)",
+    )
+    parser.add_argument(
+        "--order-size",
+        type=float,
+        default=float(os.getenv("ORDER_SIZE", "5")),
+        help="Order size (default: 5)",
+    )
+    parser.add_argument(
+        "--max-delta",
+        type=float,
+        default=float(os.getenv("MAX_DELTA", "20")),
+        help="Maximum delta (default: 20)",
+    )
+    parser.add_argument(
+        "--interval",
+        type=float,
+        default=float(os.getenv("CHECK_INTERVAL", "5")),
+        help="Check interval in seconds (default: 5)",
+    )
+    parser.add_argument("--duration", type=int, default=None, help="Duration in minutes")
+    return parser.parse_args()
+
+
+def main() -> int:
+    load_dotenv()
+    args = parse_args()
+
+    if not args.market_id and not args.slug:
+        logger.error("Provide --market-id or --slug")
+        return 1
+
+    try:
+        exchange = create_exchange(args.exchange)
+    except ValueError as e:
+        logger.error(str(e))
+        return 1
+
+    logger.info(f"\n{Colors.bold('Exchange:')} {Colors.cyan(args.exchange.upper())}")
+
+    market_id: Optional[str] = args.market_id
+    if not market_id and args.slug:
+        market_id = find_market_id(exchange, args.slug, args.market_index)
+        if not market_id:
+            return 1
+
+    strategy = AvellanedaStoikovStrategy(
+        exchange=exchange,
+        market_id=market_id,
+        gamma=args.gamma,
+        k=args.k,
+        time_horizon_hours=args.time_horizon,
+        volatility_window=args.vol_window,
+        min_spread=args.min_spread,
+        max_position=args.max_position,
+        order_size=args.order_size,
+        max_delta=args.max_delta,
+        check_interval=args.interval,
+    )
+
+    try:
+        strategy.run(duration_minutes=args.duration)
+    except KeyboardInterrupt:
+        pass
+
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())