Backtest Results

USDJPY Triple-Swap Wednesday

Dukascopy Tick Data · OOS 2022–2026 · IS 2019–2022 · commission included

The folklore 'triple swap' strategy - hold long USDJPY across the Wednesday rollover to collect three nights of positive carry in a few hours of market exposure - tested with REAL native swaps, which almost no retail backtest models. The carry arrived exactly as advertised, out-of-sample too. The strategy still lost heavily: a few hours of price exposure costs far more than three nights of swap pays.

USD/JPY (EntryHourUTC=23, ExitMinutesAfterRollover=30, FixedSLPips=40)

Carry real, killed by price exposure
$10k → $8.0k
-$2,035 net loss
Win Rate: 39.42% · PF: 0.595 · Max DD: 21.76%

USD/JPY (EntryHourUTC=23, ExitMinutesAfterRollover=30, FixedSLPips=20)

Collapsed out-of-sample
$10k → $6.4k
-$3,622 net loss
Win Rate: 35.1% · PF: 0.576 · Max DD: 38.51%

USD/JPY (EntryHourUTC=23, ExitMinutesAfterRollover=170, FixedSLPips=30)

Collapsed out-of-sample
$10k → $7.3k
-$2,675 net loss
Win Rate: 42.79% · PF: 0.681 · Max DD: 28.18%

Strategy Rules

On Wednesday at EntryHourUTC:00 (first M5 bar at/after the hour), open a long USDJPY position at market
One trade per week (Wednesday only - the triple-swap night)
Position MUST be open across the 00:00 UTC rollover into Thursday to collect the tripled swap
None - short USDJPY pays swap; the carry only exists on the long side
exit at ExitMinutesAfterRollover past 00:00 UTC Thursday - the time exit IS the strategy's exit; TP is a formality, SL is disaster protection
before coding, verify SYMBOL_SWAP_ROLLOVER3DAYS on USDJPY_CUSTOM is Wednesday; if the symbol is configured differently, the entry day input must follow the symbol setting, not the calendar assumption
no spread filter in tester (spread lives in the fills; note rollover-time spread widening is exactly what this test must survive)

Backtest Parameters

In-Sample Window2019.01.01 – 2022.01.01
Out-of-Sample Window2022.01.01 – 2026.01.01
Data SourceDukascopy tick data (real ticks)
Starting Capital$10,000
Risk per Trade1.0% of equity
Commission$3.50/side/lot (IC Markets Raw, modeled)
OOS Runs Consumed3

Methodology

Genetic optimization on the in-sample window with a net-of-costs robustness criterion; parameter-plateau selection (top candidates must have profitable neighborhoods, not just a best pass); out-of-sample window run once per frozen candidate on real ticks; verdicts gated on net profit factor, drawdown, and IS-to-OOS retention.

Key Takeaway

Two findings. First, the carry mechanism is real and our tester charged it natively: every Wednesday-crossing hold collected about 30 points (3x the +10.064-point nightly long swap), in-sample and out-of-sample alike. Second, the in-sample profits (PF 1.5-1.85 on the 23:00-entry configs) were never carry - they were a 2019-2021 early-Tokyo price drift that inverted after 2022: out-of-sample all three frozen candidates collapsed (PF 0.58-0.68, drawdowns up to 38.5%), failing 8 of 12 gates each, with the swap STILL positive (+1,305 to +2,010) but the price leg losing 3-5x that in the BOJ-intervention era. The robustness tooling flagged the fragility before the freeze (no parameter plateau); the out-of-sample confirmed it. Swap-harvesting schemes with market exposure are regime bets wearing a carry costume.

Disclaimer: Past performance is not indicative of future results. These backtest results are based on historical data with realistic commission assumptions. Real trading involves additional risks including execution delays, variable spreads, and emotional decision-making.