Multi-Strategy Crypto Trading Bot

A modular, live-trading multi-strategy engine for USD-quoted spot crypto, with:

• Robust signal ensemble (trend, momentum, mean-reversion, AR/ARMA, Fourier/seasonality, beta, VWAP/CLV).
• Exchange-agnostic market data via ccxt.
• Live order routing to Roostoo mock/live API with lot-size rounding and safety checks.
• Simple dry-run mode, CSV trade logs, and backtesting notebooks.

⚠️ Disclaimer: This software is for educational use. Crypto trading is risky. Use at your own risk.

---

Repository Structure

.
├─ deployed_bot.py           # Live trading entrypoint (dry run or live to Roostoo)
├─ backtester.ipynb          # Backtesting / experiments (notebook)
├─ multistrat_engine.ipynb   # Strategy engine exploration (notebook)
├─ Final Model.ipynb         # Final modeling / research notes (notebook)
├─ whale_supply/             # Whale supply signals / utilities (placeholder folder)
└─ README.md                 # You are here

---

What It Does

Live data

• Pulls OHLCV from the configured ccxt exchange (default: coinbase) at 5m resolution for a configurable USD universe.

Signal ensemble

The bot builds an ensemble from multiple strategies and converts it into long-only target weights:

• Donchian_Breakout (panel) — Breaks out above N-bar highs/lows on resampled bars (default: 1H / N=72).
• Cross_Sectional_Momentum (panel) — Ranks symbols by lookback returns, filters by dollar volume, buys the top fraction.
• MeanReversionStrategy — Z-score of price vs. rolling mean; enters below z_entry, exits over z_exit.
• MovingAverageStrategy — Majority vote over EMA/SMA/WMA/DEMA/TEMA/TRIMA/KAMA short vs. long crossovers.
• BreakoutStrategy — Price vs. MA bands + N-day high/low + ATR smoothing.
• BetaRegressionStrategy — Rolling beta/corr/z-score between low/high series.
• VWAPStrategy — Deviations from rolling VWAP with volume weighting.
• CLV — Close Location Value classifier from -1..+1.
• AutoRegressiveStrategy / ARMAStrategy — Residual-sign classifier from AR(p)/ARMA(p,q).
• FourierTransform — DFT + band-pass/energy/seasonality tests; constructs a weight from low/mid/high energy and recent slopes.

Each strategy contributes a [0..1] signal; the engine combines them by weighted sum, normalizes to weights, then exponential-smooths them over time.

Portfolio + execution

• Computes current weights from Roostoo balances (USD + coin holdings × last price).

• Compares to target; if L1 distance ≥ threshold, computes trades.

• Safety rails:

  • Long-only (no shorts).
  • Don’t sell more than you own.
  • Skip notionals < MIN_NOTIONAL.
  • Respect exchange lot sizes (step sizes) with rounding down.

• Sends MARKET orders to Roostoo with signed requests (HMAC-SHA256).

• Writes trades to trade_log.csv.

---

Quickstart

1) Python & dependencies

• Python 3.11+ recommended (works on 3.13 as well).
• TA-Lib requires system libs.

# Linux/macOS (example)
python -m venv .venv && source .venv/bin/activate
pip install --upgrade pip
pip install numpy pandas matplotlib requests ccxt statsmodels scikit-learn scikit-optimize scipy quantstats

# TA-Lib (choose one, OS-specific)
# macOS (brew):  brew install ta-lib
# Ubuntu:        sudo apt-get install -y ta-lib
# then:
pip install ta-lib

2) Environment variables (Roostoo)

Provide your Roostoo API credentials via env vars or in code:

export ROOSTOO_API_KEY="your_api_key"
export ROOSTOO_SECRET="your_secret"

In deployed_bot.py, the client reads:

roostoo = RoostooClient(
    api_key=ROOSTOO_API_KEY or os.getenv("ROOSTOO_API_KEY"),
    secret=ROOSTOO_SECRET or os.getenv("ROOSTOO_SECRET"),
)

3) Configure universe (optional)

Edit UNIVERSE_BASES in deployed_bot.py. Default includes:

"ZEN","AAVE","ADA","APT","ARB","AVAX","BNB","BTC","CAKE","DOGE","DOT","EIGEN",
"ENA","ETH","FET","FIL","ICP","LINK","LTC","NEAR","OMNI","ONDO","PENDLE","PENGU",
"POL","SEI","SOL","SUI","TAO","UNI","WIF","WLD","XLM","XRP","ZEC"

Make sure your ccxt exchange actually lists these as BASE/USD (e.g., ARB/USD). If not, remove/adjust symbols or swap the exchange in LiveDataPipeline(exchange_id='coinbase').

4) Dry-run or Live

In deployed_bot.py:

if __name__ == "__main__":
    DRY_RUN = False   # set True for paper trading (console prints only)
    ...

Run it:

python deployed_bot.py

• Dry run prints [DRY RUN] BUY/SELL ... without hitting the API.
• Live signs and submits orders to Roostoo’s /v3/place_order.

---

Configuration Knobs (in deployed_bot.py)

• Execution

  • MIN_NOTIONAL = 10.0 # skip tiny fills
  • THRESHOLD = 0.02 # L1 diff vs current weights to rebalance
  • LAMBDA = 0.8 # target weight smoothing (higher = more reactive)
  • sleep_seconds = 300 # loop interval

• Data

  • exchange_id='coinbase' # ccxt exchange id
  • timeframe='5m' # OHLCV interval
  • limit=1500 # bars to pull

• Lot sizes

  • RoostooClient.base_lot_sizes — per-asset step sizes (e.g., BTC=0.0001, ADA=1.0).
  • Orders are rounded down to the nearest step; 0 after rounding ⇒ skipped.

• Strategy weights

  • strategies_config = [ StrategyConfig(..., weight=..., long_only=...), ... ]
  • Combine policy: 'weighted_sum' (default) or 'mean'.

---

How It Works (Flow)

ccxt (OHLCV) ──► LiveDataPipeline ──► series_mats (open/high/low/close/vol/vwap/clv)
                                        │
                                        ▼
                             MultiStrategyEngine (panel + per-symbol strategies)
                                        │
                                        ▼
                             target weights (normalized & smoothed)
                                        │                  ▲
        Roostoo /balance ──► current weights ◄────────────┘
                                        │
                              Rebalance decision (L1 ≥ THRESHOLD)
                                        │
                                        ▼
                       Trade list with notional & lot-size rounding
                                        │
                          Roostoo /v3/place_order (signed HMAC)
                                        │
                                        ▼
                               trade_log.csv (append-only)

---

Notebooks

• backtester.ipynb Load historical OHLCV (via ccxt or your data), reproduce the series_mats dict, call the MultiStrategyEngine to compute signals and simulate PnL. Use quantstats for tear-sheets.

• multistrat_engine.ipynb Isolated strategy exploration: tune parameters, visualize signals, combine policies, sensitivity analysis.

• Final Model.ipynb Final exploration / results summary (use as research log).

Tip: When backtesting, keep strategy logic deterministic and side-effect free. The ensemble already expects per-symbol Series and panel DataFrames.

---

Whale Supply

whale_supply/ is a staging area for whale inflow/outflow or net-position metrics (e.g., from Horus/Glassnode-like sources). Integrate these as an additional strategy by:

1. Building a per-symbol time series, aligned to your OHLCV index.
2. Adding a new StrategyConfig(YourWhaleStrategy(), ['whale_signal'], weight=..., long_only=...).
3. Merging the column into series_mats (e.g., extend LiveDataPipeline to fetch it).

---

Common Errors & Fixes

• [RoostooClient] WARNING: API key/secret not set. Export ROOSTOO_API_KEY / ROOSTOO_SECRET or hard-code them for testing.

• "quantity step size error" (Roostoo response) Increase precision or adjust base_lot_sizes for that symbol. The client rounds down; if rounding→0, trade is skipped.

• No trades placed

  • Signals may all be zero (ensemble yields no allocation).
  • L1 diff below THRESHOLD.
  • Notional per order < MIN_NOTIONAL.
  • Your universe has symbols not supported by the exchange.

• TA-Lib install issues Install system libraries first (brew/apt), then pip install ta-lib.

• ccxt symbol not found Check actual symbols on your exchange; adapt to BASE/USD pairs that exist. Some venues use USDT or different tickers.

---

Extending the System

• Add a new strategy Implement a class with a signal(...) -> DataFrame|Series returning a column 'value' in [0..1] (or any real, the engine clips long-only). For panel strategies, set requires_panel = True and implement:

  def signal(self, index, open, high, low, close, volume, vwap, clv) -> pd.DataFrame

  Return a DataFrame with symbol columns → integer/float picks or weights.

• Change combine policy In MultiStrategyEngine(combine='mean'|'weighted_sum').

• Risk overlays Add volatility targeting, drawdown caps, or cash buffers at the rebalance step before order creation.

---

Logging

• Trades appended to trade_log.csv:

  timestamp,coin,side,qty,price
  2025-...Z,BTC,BUY,0.005,105000
  

• Console prints:

  • Target/current weights, L1 diff, per-order results:

    • ORDER OK ... on success.
    • ROOSTOO ERROR ... with API response.
    • SKIP ... when below step size or notional.

---

Citation / Credit

Built with ❤️ on top of:

• ccxt for market data
• TA-Lib for indicators
• statsmodels
• scikit-learn
• scikit-optimize
• quantstats