Strategy audit: Gotobi Tokyo Fix on EUR/JPY
This is what you get back from an audit. It uses one of our own strategies, already public on its results page, so every number here can be checked. Your own report is private to you: nothing about your strategy is published without your written permission.
A real edge, but a shrinking one.
- Passed 14 of 14 gates on data it had never seen (2022.01.01 – 2026.01.01).
- +$2,625 on $10,000 over 283 trades after commission and swap, profit factor 1.42, worst drawdown 6.0%.
- Every year was profitable, but 2024 (+$234) and 2025 (+$205) earned a fraction of 2022 (+$1,317).
- Our advice: run it on a demo account before any real money, and watch whether the recent years keep shrinking.
1. What we tested
Your rules, written out exactly as the EA follows them:
- Trade only gotobi days: calendar dates divisible by 5, shifted to the preceding business day when they fall on a weekend (Japanese settlement convention)
- Enter long EUR/JPY at 15:55 UTC the day before the fix (nine hours before the 9:55 JST Tokyo fix; JST has no DST, so the UTC clock is constant)
- Hard exit at the fix, 00:55 UTC, regardless of profit or loss - the flow ends at the fix
- Catastrophe stop 60 pips; take-profit 200 pips (a formality - never reached); skip entry if the spread exceeds 3 pips
- One trade per gotobi day, long only, 1% of equity risked per trade
- Parameters are the USDJPY parent's frozen candidate c2, inherited unchanged - this pair was never optimized on
| In-Sample Window | 2019.01.01 – 2022.01.01 |
| Out-of-Sample Window | 2022.01.01 – 2026.01.01 |
| Data Source | Dukascopy tick data (real ticks) |
| Starting Capital | $10,000 |
| Risk per Trade | 1.0% of equity |
| Commission | $3.50/side/lot (IC Markets Raw, modeled) |
| Swap | Charged natively, but at one fixed rate for the whole window (MT5 applies the current rate backward). Worth 10% of net profit; with swap zeroed the OOS window is +23.6% at PF 1.37. |
| OOS Runs Consumed | 1 |
Settings are fixed on the first period. The second period is run once, at the end, with nothing changed. In this case no settings were fitted to EUR/JPY at all: they were carried over unchanged from the USD/JPY version.
2. Results on unseen data
3. The scorecard
The same 14 pass-or-fail gates we use on every published result. A strategy has to pass all of them to be called a real edge.
Scorecard: our 14 gates
| Gate | This strategy | Pass rule | |
|---|---|---|---|
| ✓ | Enough in-sample trades | 213 | at least 150 |
| ✓ | Enough out-of-sample trades | 283 | at least 100 |
| ✓ | Trades often enough | 71 a year | at least 30 a year |
| ✓ | In-sample profit factor | 1.56 | at least 1.30 |
| ✓ | Out-of-sample profit factor | 1.42 | at least 1.15 |
| ✓ | Out-of-sample drawdown | 6.0% | no more than 20% |
| ✓ | Return per trade vs its swings | 0.129 | at least 0.05 |
| ✓ | Edge kept out-of-sample | 90.9% | at least 70% |
| ✓ | Return kept out-of-sample | 143.5% | at least 50% |
| ✓ | Drawdown did not blow up | 1.21x in-sample | at most 1.5x |
| ✓ | Profitable most years | 100% | at least 50% |
| ✓ | Survives higher costs | +$1,824 | still profitable |
| ✓ | Not a lucky order of trades | 1.4% chance of loss | at most 15% |
| ✓ | Beats doing nothing clever | 1.22x the best baseline | at least 1x |
Scored with today's rules. What each gate means: How we test.
4. Costs
- Spread: from real tick data, so it widens at news and at night exactly as it did in the market.
- Commission: $3.50 per lot per side, an IC Markets Raw account.
- Swap: charged on every overnight hold, including the triple charge on Wednesdays.
- Stress test: with an extra $5 commission per lot per side the strategy still made +$1,824, profit factor 1.27.
5. Luck check (Monte Carlo)
We reshuffled and resampled the 283 trades 10,000 times to see how much of the result could be a lucky run.
The actual drawdown (6.0%) sat in the middle of the shuffled range, so the test period was not unusually kind. Size your risk for the 9.3%, not the 6.0%.
6. Does it beat the simple alternative?
Return divided by worst drawdown, over the same years, against two strategies that need no edge at all:
| Return a year | Worst drawdown | Return / drawdown | |
|---|---|---|---|
| This strategy | 6.0% | 6.0% | 0.99 |
| Buy and hold EUR/JPY | 8.9% | 10.9% | 0.81 |
| 12-month trend following | 3.5% | 12.2% | 0.29 |
1.22x the best alternative, during one of the biggest falls in the yen in decades, when simply holding EUR/JPY did very well.
7. Does the EA follow your rules?
We check the trades the EA actually placed against the rules: risk per trade, stop and target distance, holding time, entry window, one trade at a time, long only, the right symbol and timeframe. All 10 checks passed. A strategy that looks good because the code quietly breaks its own rules fails here.
8. What we would tell you
- The edge is real on unseen data and survives higher costs. It is not a fluke of trade order.
- It is getting weaker. The last two years earned far less than the first two.
- Next step: a forward test on a demo account. We run this one live on our forward test; the forward test, not this backtest, decides whether it is still alive.
- A backtest describes the past. It is not a promise of future returns and not trading advice.
Most strategies we test fail, and the report says so plainly. Join the waitlist to hear first when audit slots and the price are set.
Join the waitlist