Strategy idea: Pattern similarity-based signals via Chart Library API

[grahammccain]

Idea

Backtesting.py makes it easy to test indicator-based strategies. Here's a strategy concept that uses historical pattern similarity as the signal: find the most similar past chart patterns and trade based on their aggregate forward returns.

Chart Library has 24M+ pre-computed embeddings across 19K US stocks (10 years). Its API returns the top matches with actual forward returns for any (symbol, date).

Example strategy

from backtesting import Backtest, Strategy
import requests

class PatternSimilarity(Strategy):
    """
    Buy when historically similar patterns had positive forward returns.
    Sell when they had negative forward returns.
    """
    lookback_threshold = 0.6  # min win rate to enter long
    
    def init(self):
        self.pattern_signals = {}
        
    def next(self):
        date_str = str(self.data.index[-1].date())
        
        if date_str not in self.pattern_signals:
            try:
                resp = requests.get("https://chartlibrary.io/api/v1/search", 
                    params={"symbol": "AAPL", "date": date_str, "timeframe": "RTH"},
                    headers={"X-API-Key": "your-key"}, timeout=5)
                matches = resp.json()["matches"]
                win_rate = sum(1 for m in matches if m["return_5d"] > 0) / len(matches)
                avg_return = sum(m["return_5d"] for m in matches) / len(matches)
                self.pattern_signals[date_str] = {
                    "win_rate": win_rate, "avg_return": avg_return
                }
            except Exception:
                return
        
        sig = self.pattern_signals.get(date_str)
        if not sig:
            return
            
        if not self.position and sig["win_rate"] >= self.lookback_threshold:
            self.buy()
        elif self.position and sig["win_rate"] < 0.4:
            self.position.close()

For production, you'd pre-fetch all signals into a DataFrame column rather than calling the API in the loop.

What the API provides per match

• Symbol, date, timeframe of the historical match
• L2 distance (similarity score)
• Forward returns: 1-day, 3-day, 5-day, 10-day

Details

• Free: 200 calls/day, no credit card
• Docs: https://chartlibrary.io/developers
• Coverage: 19K US equities, 10 years, 8 timeframes

This is essentially a "nearest neighbor" trading strategy — find the most similar historical charts and bet on the average outcome.