🏔️ Vanta Crunch — Challenge Package

Predict short-term crypto returns. Beat the leaderboard.

Quickstart

pip install vanta-challenge matplotlib  # matplotlib optional, for plots

1. Write a model

from vanta.tracker import TrackerBase
from vanta.indicators import RSI, EMA, OrderBookImbalance, FundingSignal

class MyModel(TrackerBase):
    def __init__(self):
        super().__init__()
        self.rsi = RSI(period=14)
        self.ema_fast = EMA(period=5)
        self.ema_slow = EMA(period=20)
        self.obi = OrderBookImbalance(period=10)
        self.funding = FundingSignal(period=8)

    def tick(self, data: dict):
        super().tick(data)
        for candle in data.get("candles_1m", []):
            price = float(candle["close"])
            self.rsi.update(price)
            self.ema_fast.update(price)
            self.ema_slow.update(price)

        # Order book imbalance (if available)
        ob = data.get("orderbook")
        if ob:
            self.obi.update(ob["imbalance"])

        # Funding rate (if available)
        fund = data.get("funding")
        if fund:
            self.funding.update(fund["funding_rate"])

    def predict(self, symbol: str, horizon_seconds: int, step_seconds: int) -> dict:
        if not self.rsi.ready:
            return {"expected_return": 0.0}

        signal = 0.0

        # RSI + trend filter
        if self.rsi.value < 30 and self.ema_fast.value > self.ema_slow.value:
            signal += 0.003
        elif self.rsi.value > 70 and self.ema_fast.value < self.ema_slow.value:
            signal -= 0.003

        # Order book pressure boost
        if self.obi.ready:
            signal += self.obi.value * 0.002  # imbalance [-1,1] → ±0.002

        # Funding rate mean-reversion (negative = contrarian)
        if self.funding.ready:
            signal -= self.funding.value * 50  # e.g. 0.0001 × 50 = 0.005

        return {"expected_return": max(-0.01, min(0.01, signal))}

2. Backtest it

from vanta.backtest import BacktestRunner

result = BacktestRunner(model=MyModel()).run(
    subject="BTCUSDT",
    start="2025-06-01",
    end="2026-01-01",
)
result.summary()  # formatted metrics table
result.plot()     # equity curve, drawdown, hit rate, scatter

3. Compare with other strategies

from vanta.backtest import compare
from vanta.examples import MeanReversionTracker, TrendFollowingTracker

compare([
    ("Mean Reversion", MeanReversionTracker()),
    ("Trend Following", TrendFollowingTracker()),
    ("My Model", MyModel()),
], start="2025-06-01", end="2026-01-01")

4. Sweep parameters

from vanta.backtest import sweep

sweep(
    model_fn=lambda period: MyModel(period=period),
    params={"period": [7, 14, 21, 28]},
    start="2025-06-01", end="2026-01-01",
)

How It Works

• Data: Multi-timeframe OHLCV candles (1m, 5m, 15m, 1h) + order book depth + funding rates from Binance
• Prediction: Every 15 minutes, predict the 1-hour forward return
• Scoring: score = expected_return × actual_return
• History: ~1 year of data available for backtesting (Jan 2025 → present)

Scoring

score = expected_return × actual_return

Direction AND magnitude matter. See docs/scoring-guide.md for details.

Indicators

Built-in stateful indicators — no pandas/numpy required:

from vanta.indicators import SMA, EMA, RSI, MACD, BollingerBands, ATR, VWAP, Returns

rsi = RSI(period=14)
rsi.update(price)     # feed one value
rsi.value             # current reading (or None if not ready)
rsi.ready             # True once enough data accumulated

Examples

Seven starter strategies in vanta/examples/:

Docs

• Scoring Guide — how scoring works, what the leaderboard metrics mean
• Strategy Guide — practical tips, common approaches, pitfalls
• Compare Notebook — run all strategies side by side

Model Interface

class TrackerBase:
    def tick(self, data: dict) -> None:
        """Called with each new data batch.
        data = {
            "symbol": "BTCUSDT",
            "asof_ts": 1234567890,

            # Multi-timeframe OHLCV candles
            "candles_1m":  [{"ts": ..., "open": ..., "high": ..., "low": ..., "close": ..., "volume": ...}, ...],
            "candles_5m":  [...],   # last 60 bars (5h)
            "candles_15m": [...],   # last 40 bars (10h)
            "candles_1h":  [...],   # last 24 bars (1 day)

            # Order book snapshot (or None if unavailable)
            "orderbook": {
                "best_bid": 50000.0, "best_ask": 50001.0,
                "spread": 1.0, "mid_price": 50000.5,
                "bid_depth": 123.4, "ask_depth": 98.7,
                "imbalance": 0.11,  # (bid-ask)/(bid+ask), range [-1, 1]
                "bids_top": [[50000.0, 1.5], ...],
                "asks_top": [[50001.0, 1.2], ...],
            },

            # Funding rate / basis (or None if unavailable)
            "funding": {
                "funding_rate": 0.0001,   # current 8h funding rate
                "mark_price": 50000.5,
                "index_price": 50000.0,
                "basis": 0.00001,          # (mark-index)/index
                "next_funding_ts": 1234571490,
            },
        }
        """

    def predict(self, symbol: str, horizon_seconds: int, step_seconds: int) -> dict:
        """Return your prediction.
        Returns: {"expected_return": float}
        """