Same BatchBacktesting engine as the earlier indicator experiment, but emphasized at portfolio scale: many tickers, two strategies, equities plus crypto, aggregate winners and losers. Version française.
Why run many tickers at once
Single-ticker backtests flatter you. A grid run answers:
- Does this rule only work on one heroic symbol?
- Are disasters clustered in a sector or era?
- Is crypto dominating the leaderboard because volatility is unbounded in the sim?
Architecture (same as BatchBacktesting)
| Layer | Role |
|---|---|
| Data adapters | FMP + Binance fetch |
| Strategy objects | EMA, MACD parameter sets |
| Runner | Thread pool, CSV output, optional plots |
output/ | Per-run artifacts — not committed to GitHub by default |
Pre-calculated results are not in the upstream repo (too user-specific). You generate locally, then inspect output/charts/....
Running at scale
run_backtests(get_SP500(), strategy=EMA, num_threads=12, generate_plots=True)
run_backtests(get_SP500(), strategy=MACD, num_threads=12, generate_plots=True)
run_backtests(get_all_crypto(), strategy=EMA, num_threads=12, generate_plots=True)
run_backtests(get_all_crypto(), strategy=MACD, num_threads=12, generate_plots=True)
Tune num_threads to your machine and API rate limits.
Reading leaderboard extremes (EMA example)
High returns often mix:
- Long bullish windows on leveraged crypto pairs.
- Mean-reversion false signals in sideways stocks (sometimes lucky).
- Parameters tuned implicitly by the default strategy class.
Deep negatives flag:
- Illiquid symbols where the sim still trades.
- Trend rules in collapsing sectors.
- Stablecoins or broken pairs treated like normal tickers.
Before narrating “MACD beats EMA,” compare median return, not only top decile.
Honest limitations
- Not financial advice.
- Fees, borrow, halts, and corporate actions are only as good as the data feed + library defaults.
- Past batch output ≠ future edge.
Takeaway
Use multi-ticker backtests as smoke tests for rules — find where they break, not where to deploy capital.
Related posts
Originally published on Medium. Full code listings: BatchBacktesting.