update 03-29 01:36

src/config.py

@@ -65,6 +65,8 @@ class RiskConfig:
 class FeesConfig:
     taker_fee_5m: float = 0.0156
     taker_fee_15m: float = 0.03
+    fee_rate: float = 0.045
+    fee_exponent: int = 1
 
     def fee_for_timeframe(self, timeframe: str) -> float:
         """Return the taker fee for a given timeframe string (e.g. '5M')."""
@@ -76,6 +78,13 @@ class FeesConfig:
             raise ValueError(f"Unknown timeframe '{timeframe}'; expected one of {list(mapping)}")
         return mapping[timeframe]
 
+    def dynamic_fee(self, price: float, shares: int = 1) -> float:
+        """Compute Polymarket's dynamic taker fee for a given price and share count.
+
+        Formula: shares * price * fee_rate * (price * (1 - price)) ** fee_exponent
+        """
+        return shares * price * self.fee_rate * (price * (1.0 - price)) ** self.fee_exponent
+
 
 @dataclass(frozen=True)
 class BinanceConfig:

src/strategy/temporal_arb.py

@@ -109,7 +109,8 @@ class TemporalArbStrategy:
         )
 
         # 6. Edge calculation (after fees)
-        taker_fee = self.fees.fee_for_timeframe(timeframe)
+        # Use dynamic fee formula: fee per share / price = effective fee rate
+        taker_fee = self.fees.dynamic_fee(poly_price, 1) / poly_price
         edge = estimated_prob - poly_price - taker_fee
 
         if edge < self.arb.min_edge:
@@ -118,7 +119,7 @@ class TemporalArbStrategy:
         # 7. Position sizing
         asset = Asset(symbol)
         size = self.calculate_kelly_size(
-            edge=edge,
+            estimated_prob=estimated_prob,
             price=poly_price,
             balance=self.balance,
             max_size=self.risk.max_position_per_market_usd,
@@ -181,25 +182,25 @@ class TemporalArbStrategy:
         """
         abs_change = abs(price_change_pct)
 
-        # Factor 1: Base probability from price magnitude
-        # 0.15% → ~70%, 0.3% → ~82%, 0.5% → ~90%, 1.0%+ → ~95%
-        base_prob = 0.55 + abs_change * 1.0  # 1.0 scaling factor
-        base_prob = min(base_prob, 0.95)
+        # Factor 1: Sigmoid base probability from price magnitude
+        # k=18.0 steepness, inflection at 0.10% change
+        # 0.03% → ~0.53, 0.10% → ~0.65, 0.20% → ~0.80, 0.50% → ~0.93
+        k = 18.0
+        threshold = 0.10
+        sigmoid_val = 1.0 / (1.0 + math.exp(-k * (abs_change - threshold)))
+        base_prob = 0.52 + sigmoid_val * 0.43  # maps [0,1] → [0.52, 0.95]
 
         # Factor 2: Time decay — more time elapsed = more confirmation
         # If 80% of window has passed and price is still in this direction,
         # the probability of reversal is lower
-        elapsed_fraction = max(0, 1.0 - (time_remaining / total_window_sec))
-        # Sigmoid-like boost: ramps up in the last 40% of the window
+        elapsed_fraction = 1.0 - (time_remaining / total_window_sec) if total_window_sec > 0 else 0.5
+        # Ramps up in the last 40% of the window
         time_factor = 1.0 + 0.08 * max(0, elapsed_fraction - 0.6) / 0.4
         time_factor = min(time_factor, 1.08)
 
         # Factor 3: Volatility / momentum strength
         # Very large moves (>0.5%) get extra confidence
-        if abs_change > 0.5:
-            vol_boost = min(0.05, (abs_change - 0.5) * 0.1)
-        else:
-            vol_boost = 0.0
+        vol_boost = min(0.05, (abs_change - 0.5) * 0.1) if abs_change > 0.5 else 0.0
 
         final_prob = base_prob * time_factor + vol_boost
         return min(0.95, max(0.50, final_prob))
@@ -210,7 +211,7 @@ class TemporalArbStrategy:
 
     def calculate_kelly_size(
         self,
-        edge: float,
+        estimated_prob: float,
         price: float,
         balance: float,
         max_size: float,
@@ -224,7 +225,7 @@ class TemporalArbStrategy:
             return 0
 
         b = (1.0 / price) - 1.0  # Payout odds
-        p = edge + price          # estimated_prob (edge = prob - price - fee)
+        p = estimated_prob
         q = 1.0 - p
 
         if b <= 0: