Find the relationship.
Test the reversion.
Rolling Engle–Granger screen, USD-notional hedge sizing, next-bar simulation and transparent two-leg costs. No fabricated performance.
Configure the experiment
Free hourly historical forex data. Market History provides UTC H1 bid quotes (last five years, keyless); AlgoSpecial provides broker-clock H1 history. Both require modeled execution spreads unless you import observed bid/ask CSVs.
Dates below define the OUT-OF-SAMPLE TEST. The downloader automatically retrieves the 180-day training history before the start date.
Upload two H1 CSV files with timestamp,open,high,low,close. For stronger execution realism, also upload each pair's ASK candle CSV; the main CSV should be BID. Files must share the same timestamp clock. Dukascopy supplies bid and ask history.
Execution costs and assumptions expand
Spread figures are full bid–ask spreads per pair, paid across round trip; not 0.9 pips each side. If paired bid/ask CSVs are supplied, observed bid/ask prices replace the modeled spread for that pair. Financing is a configurable drag, not actual broker swap history.
Walk-forward performance
Realized equity curve
At closed positionsExecution breakdown
Rolling cointegration / regime diagnosis
Train only · Monthly refits| Test window | Corr. | Hedge β | ADF screen t | Half-life | Status | Trades | Test return |
|---|
ADF screening critical value is approximate for a two-variable Engle–Granger residual test. Correlation measures return co-movement, not cointegration. Training uses prior completed H1 bars only.
Two-leg trade ledger
| Entry (source clock) | Direction | EUR units | GBP units | Hold | Gross P&L | Execution + finance | Net P&L | Exit reason |
|---|
Data integrity & methodological limitations
This is an educational research simulator. Spread convergence is not guaranteed. Two-leg forex trades have execution, financing and residual currency risk. Alpha requires robust out-of-sample and forward validation.