Yen Regime Overnight
- Experts
- Version: 1.0
- Activations: 10
Yen Regime Overnight - an overnight strategy on the Japanese yen that is fitted
to the 2021-2026 yen regime, says so on this page, and pauses itself when that
regime ends.
No grid, no martingale, no averaging down, hard SL on every trade.
Symbols: USDJPY and EURJPY. Risk per trade: 0.75% of equity, capped at 2.0%.
Every business evening the EA buys USDJPY and EURJPY at 23:00 Japan time and
closes both at 09:53 next morning. Long only - short the yen overnight, every
night, collecting the interest-rate differential while it holds. One position per
symbol, a hard stop at 0.5 x ATR(14) of the daily chart, no take profit, no
trailing stop. 392 trades a year is the two symbols added together (196 each);
84% end at the time exit, 16% on the stop.
== THE FIT WINDOW: WHAT WE OPTIMISED, AND WHAT IT SCORED ==
Entry time, exit time and stop multiple were chosen by measuring 640 combinations
on August 2021 - August 2026, so the figures below were measured on the data that
selected them. That is an in-sample result.
USDJPY + EURJPY, 0.75% risk, compounded on one account, regime filter on (the
default). Tick model: Model 1 (ticks generated from M1 bars). The two symbols are
not one tester run; they are two trade lists combined on one account.
1,978 trades (392 a year, both symbols). CAGR 20.5%, 2.57x over 5.04 years
Profit factor 1.27. Win rate 56.1%. Average win / average loss 0.99
Maximum drawdown 13.2% at 0.75% risk, 8.88% at 0.50%, 17.5% at 1.00%
Longest losing streak 15. Longest flat period 652 days (1.79 years)
Best year 2022 +49.65%. Worst full year 2023 -4.56%. 2024 only +1.36%
Real ticks (Model 4) exist only for August 2024 - August 2026, about 2 years, and
52 whole days of USDJPY are missing inside it (90.2% tick coverage), so "100%
real ticks" is not something we can write. That window - 732 trades, 1.25, 8.71%
drawdown - is reference only; generated ticks flatter the result by about 1.8%.
== THE WHOLE RECORD, JANUARY 1999 - AUGUST 2026 ==
Whole record: 8,125 trades, profit factor 1.11, drawdown 44.4%, CAGR 6.6%,
5.28x, and 2,807 days (7.69 years) without a new equity high
1999-2008: profit factor 0.934, drawdown 37.7%, 0.71x - 1999-2008 lost money,
and that is with the filter running
2009-2019: 1.16, drawdown 19.1%. 2020-2026: 1.22, drawdown 13.2%
15 of the 26 years that traded were positive. Worst year 2007, -27.36%
Of the nine five-year windows from 1999-2003 to 2007-2011, eight are below 1.00,
the worst 0.696 (the ninth, 2002-2006, is 1.011); all 24 are in the risk
disclosure. Best 1%
of trades removed (81 trades) leaves 1.02: a strategy that works in one kind of
market and waits through the others.
== THE REGIME FILTER - THE PART THAT DECIDES WHEN TO STOP ==
The EA measures the 12-month change in USDJPY (252 daily bars). Below -10% it
stops opening new positions; above 0% it starts again. Open positions are never
touched. On by default. Over 27.6 years it fired 16 times and held the strategy
out of the market 28.1% of the time: 41.1% of 1999-2008, 30.9% of 2009-2019, only
4.0% of 2020-2026.
Cost in a good regime: in the fit window it declined 112 signals of 2,090 (5.4%),
profit factor rose 1.2660 to 1.2708, drawdown was identical at 13.2%, CAGR fell
0.93 points. Worth in a bad one, whole record, off against on: 1.03 against 1.11,
drawdown 83.0% against 44.4%, longest flat period 8,540 days (23.4 years) against
2,807, equity 1.85x against 5.28x. With it off the account falls to 0.182x by
2010 and regains its starting equity in May 2022. Being able to switch the filter
off is not the same as being able to run it off.
It is a brake, not an engine: 1999-2008 with the filter on still ends at 0.71x, a
0.34x decade made survivable, not profitable. Started 1 January 2007 with no
filter an account was -43.3R six years later; the filter paused after 12.3 months
at -22.9R. In 1999 and in 2011 it produced an entire calendar year with no trades
at all.
The filter looks up a symbol named USDJPY. If your broker names it USDJPY.m or
similar it cannot load the history and switches itself off silently, while
trading continues unprotected. The EA says so in the log, on the panel and in the
tester summary; one input fixes it.
== SWAP, RISK PER TRADE, AND WHY TWO SYMBOLS IS NOT RISK SPREAD ==
This EA is short the yen every night, so it receives swap. Fit window: swap is
15.0% of net profit,
and stripping it out leaves profit factor 1.21 with a 16.0% drawdown. Whole
record: 31.8%. USDJPY alone across the whole record: 58.1% (EURJPY 8.6%). We do
not claim this strategy is independent of carry - our other yen product can make
that claim, this one cannot.
Risk is 0.75% of equity per trade, per symbol - the figure every drawdown above
assumes. Return and drawdown move together, and lowering risk raises the account
you need, in exact proportion. Nothing is a fixed lot: when your broker's
smallest lot would exceed your percentage, the EA declines the trade rather than
round the risk up.
Both positions were open on all 989 trading nights of the fit window - 100% - in
the same direction: 1.50% of equity at risk in a single night. Drawdown adds
rather than cancels - across the whole record USDJPY alone drew down 33.3% and
EURJPY alone 24.3%, the pair 44.4%. Two symbols here is not risk spread across
two markets.
== BEFORE YOU BUY ==
Built in: a hard stop on every position, sent with the order and verified
afterwards; the time exit; the regime filter; equity-percent sizing; spread
protection; daily loss and profit limits that watch this EA's own trades on its
own symbol - not your account as a whole; a News filter, off by default (the
tester cannot reach the MT5 calendar, and the research behind every figure here
used no filter at all); a 14-line panel with the resolved times, the regime state
and the reason for every declined signal.
Recommended 1,500 USD, minimum 1,000 USD. Account size buys signals, not
leverage; share actually taken at 0.75% risk, measured:
USDJPY EURJPY (fit window)
500 USD 39.5% 35.3%
800 USD 84.4% 78.5%
1,000 USD 96.3% 97.3%
1,500 USD 100% 100%
Across the whole record those rows read 43.5% / 22.7%, 89.7% / 75.0%,
97.5% / 90.0% and 100% / 97.7%. The arithmetic minimum is 560 USD and we will not
print that number without this table: at that level you miss roughly 60% of the
signals even inside the window we fitted. At 0.50% risk the same 1,000 USD
account takes only 69.5% and 60.2%.
Live, the EA resolves Japan time through GMT; in the tester it falls back to a
daylight-saving model (winter GMT+2, US rules by default). It prints them at
startup and on the panel - please check them against your broker's clock, because
an hour of error is a materially different strategy. If you
reproduce our backtest, avoid Wednesday: rollover is triple then, and on another
of our EAs the same period measured 1.36 one day and 1.58 on a Wednesday
afternoon.
We run this EA on our own live account (since 2026-08-31). There is no published
track record yet, no MQL5 signal and no reviews. Every number here is backtest
data.
Requirements: MetaTrader 5, a broker quoting USDJPY and EURJPY, uptime at 23:00
Japan time, one chart per symbol. Those symbols are recommendations, not
restrictions.
== WEAKNESSES, OUR OTHER EAS, AND PRICE ==
The window was chosen by optimisation. The whole record is weak. The average win
is slightly smaller than the average loss. It depends on carry. It can stop for a
year: if Japanese policy normalises and the yen reverses, the filter pauses this
EA and you own a strategy that is deliberately not trading - the design working,
and still unpleasant to own. If you want one validated across all market
conditions, our defensive line is where to look.
Gotobi Overnight JPY uses the same night window: it trades only the eves of
Japanese settlement dates, on a date effect measured across 27 years. This one
trades every business evening on the yen regime itself, declared as a fitted
window with an automatic pause.
Trading leveraged instruments carries a substantial risk of loss; past results do
not indicate future results. The user guide, the risk disclosure document and the
.set files are attached in the Comments tab. Mode presets ship inside the EA
(Standard 0.75% / Conservative 0.375%). Updates are free.
The launch price is the lowest this EA will ever sell for. It rises roughly one
step per ten copies sold and stops permanently at 199 USD - a ceiling, not a
target. A price that climbs with no stated end tells an
early buyer nothing; a stated limit is checkable.
