fx-marketslisted

# FX markets FX breaks three assumptions an equity toolchain makes: **there is no consolidated tape, the quote convention is not uniform, and a position earns or pays interest every night.** ## 1. 🚨 A spot-only backtest is missing the carry A currency position is two interest rates. Holding long AUD/short USD earns the AUD rate and pays the USD rate every night — the **swap** or **rollover**. That is not a fee; for a carry trade it is the entire thesis. ✅ Demonstrated in `scripts/fx_conventions.py`: a long AUDUSD position over the sample returns **spot only −2.97%/yr (Sharpe −0.27) — a loser** — and **total +1.34%/yr (Sharpe +0.19) — a winner**, on the same price series. **The sign of the result flips.** **If your FX backtest computes `pct_change()` on a spot series and stops, it is not a backtest of a position anyone can hold.** Add `carry_return(spot, r_base, r_quote, days)`. ✅ **Covered interest parity verified live to 0.001%** across two CME contracts. The size of the omission: leaving FX rollover out **overstates a long-EURUSD backtest by 1.30%/yr — or 3.91%/yr of equity at 3x leverage**, which is where most FX strategies actually run. ⚠️ Retail rollover is not the interbank differential — brokers mark it up, often asymmetrically, so the carry you actually receive is smaller than parity implies and the carry you pay is larger. Model the broker's published swap rates, not the policy rates, when the strategy is carry-dependent. ## 2. 🚨 Pip size is not uniform — a