# TOKYO FIX GOTOBI ANOMALY (USD/JPY) β€” EMPIRICAL QUANTITATIVE RESEARCH **Document Status:** πŸ§ͺ **EMPIRICAL RESEARCH CANDIDATE (TESTED BY ANTIGRAVITY)** **Date of Empirical Backtest:** 2026-10-10 **Tested by:** Antigravity AI [withheld: a line that refers to the lab's private live strategy] --- ## 1. Context & Microstructural Mechanics * **Origin / Inspiration:** Systematic trader Kieran Duff ([@kieran__duff](https://x.com/kieran__duff) / Darwinex PM for DARWIN `XAQP`) noted commercial settlement flow anomalies around the Tokyo Fix. * **The Gotobi Phenomenon (五十ζ—₯):** In Japan, days ending in 5 or 0 (5th, 10th, 15th, 20th, 25th, 30th of each month, or the preceding business day if falling on a weekend) are traditional commercial settlement dates. Japanese industrial importers must settle massive USD-denominated invoices (energy, oil, semiconductors) with commercial banks. * **The Tokyo Fix (Nakane - δ»²ε€€):** At **09:55 JST** (00:55 UTC / 03:55 Sofia), Japanese major banks determine and fix the official commercial foreign exchange rate for the day. * **Why Naive Retail Fails (The Front-Running Trap):** - Popular retail advice claims: *"Buy USD/JPY at 09:00 JST, sell at 09:55 JST."* - Our 11.5-year empirical data (`quantlab/data/inbox/oanda_m1/usdjpy_m1.parquet`) showed that entering long at 09:00 JST loses money net of spread ($-469$ pips, $t = -0.97$). HFTs and bank algorithmic desks front-run the retail flow between 08:00 and 08:30 JST, causing adverse selection by 09:00. * **The Structural Asymmetry (Post-Fix Demand Cliff):** - Once the clock strikes 09:55 JST, the fixing rate is permanently locked. Commercial importer demand instantly drops to zero. - Market makers who accumulated USD long inventory to service client orders dump excess inventory into the market, driving a sharp mean-reversion drop. --- ## 2. Model Architecture: Conditional Tokyo Fix Fade Tested in Python via [`prop_eur2k_lab/_test_gotobi_stretched_fade.py`](/prop_eur2k_lab/_test_gotobi_stretched_fade.py): * **Condition / Filter:** Pre-fix price expansion must be positive: $$\text{Price}_{09:55\text{ JST}} - \text{Price}_{08:30\text{ JST}} > 0\text{ pips}$$ *(Ensures commercial demand actually pushed the market up; prevents fading into a macro downtrend).* * **Entry:** Short at Bid exactly at **09:55:00 JST** (00:55 UTC / 03:55 Sofia). * **Stop Loss:** Strict fixed $SL = 20.0$ pips above entry. * **Exit:** Time-based exit at **15:00:00 JST** (06:00 UTC / 09:00 Sofia β€” Tokyo market close / European pre-open). * **Friction Applied:** Real OANDA M1 Bid/Ask spread execution PLUS an additional **0.5 pips per trade** for commissions ($5/lot) and slippage. --- ## 3. Empirical Results (USD/JPY, 2015–2026, 11.5 Years) ### A. Headline Performance Metrics | Metric | Value | Institutional Target (RULES Β§8) | Status | |---|---|---|---| | **Data Span** | 2015-01-02 to 2026-08-17 (11.5 years) | $\ge 5$ years | βœ” PASS | | **Total Trades** | 1,124 trades (~97.7 trades/year) | $\ge 100$ trades | βœ” PASS | | **Win Rate** | **56.3%** | $> 50\%$ | βœ” PASS | | **Profit Factor** | **1.92** | $\ge 1.50$ | βœ” PASS | | **Total Return** | **+326.17 R** (+28.4 R / year) | $> 0$ | βœ” PASS | | **Mean Return / Trade** | **+0.290 R** | $> +0.10$ R | βœ” PASS | | **Max Drawdown** | **12.38 R** | $< 25$ R | βœ” PASS | | **Return / MaxDD** | **26.3x** | $> 5x$ | βœ” PASS | | **Annualized Sharpe** | **2.24** | $> 1.0$ | βœ” PASS | | **t-statistic** | **+7.52** | $\ge 3.43$ (canonical) | βœ” **PASS (Extreme Statistical Significance)** | --- ### B. In-Sample vs. Out-of-Sample Split Validation * **In-Sample (2015–2020, 6 Years):** - Trades: 601 - Win Rate: 60.1% - Profit Factor: **2.27** - Total Return: **+190.36 R** - $t$-statistic: **+6.74** * **Out-of-Sample (2021–2026, 5.5 Years):** - Trades: 523 - Win Rate: 50.9% - Profit Factor: **1.66** - Total Return: **+135.81 R** (+24.7 R / year) - $t$-statistic: **+4.13** *(Passes the out-of-selection $t \ge 3.43 / 4.40$ threshold)* --- ### C. Year-by-Year Performance Breakdown (Unconditioned Exit 15:00 JST) | Year | Sessions | Net Pips (All Days) | Win Rate | $t$-stat | Gotobi Days Sub-cohort | |---|---|---|---|---|---| | **2015** | 101 | +1,015.7 pips | 65.3% | +3.43 | N=16: +202.6 pips (+12.7p/tr) | | **2016** | 101 | +1,490.0 pips | 60.4% | +3.35 | N=19: +264.4 pips (+13.9p/tr) | | **2017** | 104 | +936.3 pips | 69.2% | +4.70 | N=20: +236.6 pips (+11.8p/tr) | | **2018** | 94 | +636.4 pips | 63.8% | +3.46 | N=16: +163.7 pips (+10.2p/tr) | | **2019** | 92 | +329.3 pips | 60.9% | +2.90 | N=20: +178.2 pips (+8.9p/tr) | | **2020** | 109 | +766.4 pips | 65.1% | +3.45 | N=24: +55.3 pips (+2.3p/tr) | | **2021** | 100 | +270.7 pips | 61.0% | +1.85 | N=19: +107.8 pips (+5.7p/tr) | | **2022** | 89 | +1,458.6 pips | 65.2% | +2.85 | N=19: +734.7 pips (+38.7p/tr) | | **2023** | 95 | +424.4 pips | 60.0% | +1.16 | N=21: -26.2 pips (-1.3p/tr) | | **2024** | 85 | +1,246.1 pips | 55.3% | +2.38 | N=20: +316.0 pips (+15.8p/tr) | | **2025** | 108 | +1,746.8 pips | 68.5% | +5.09 | N=24: +814.7 pips (+34.0p/tr) | | **2026** | 46 | +514.5 pips | 60.9% | +2.37 | N=10: +55.9 pips (+5.6p/tr) | | **Total**| **1,124**| **+10,835.2 pips**| **63.3%**| **+9.54**| **Gotobi: +3,103.7 pips (+13.6p/tr, 68.0% win)** | *Observation:* **12 out of 12 years are positive.** On Gotobi settlement days specifically, the average return per trade surges by **+58%** (+13.6 pips vs +8.6 pips on non-Gotobi days) with a 68.0% win rate. --- ## 4. Cross-Asset Generalizability Tested in [`prop_eur2k_lab/_test_gotobi_multipair.py`](/prop_eur2k_lab/_test_gotobi_multipair.py): * **EUR/JPY:** Post-Fix Short to 15:00 JST is positive (+1,588.3 pips, $t = +1.36$, OOS 2023-2026 $t = +2.32$). * **GBP/JPY:** Post-Fix Short to 15:00 JST is positive (+1,341.4 pips, $t = +0.74$). * **AUD/JPY:** Negative across all horizons. *Microstructural rationale:* Australian macroeconomic releases (RBA, employment, inflation) drop between 01:30 and 02:30 UTC, directly colliding with and overpowering Tokyo Fix flows. * **Conclusion:** **USD/JPY is the pure structural expression** because the commercial flows are explicitly US Dollar settlements for Japanese energy/tech imports. --- ## [withheld: a section about the lab's private live strategy] ## 6. Authoritative Status & Validation Directives 1. **Status:** πŸ§ͺ **EMPIRICAL RESEARCH CANDIDATE (Tested by Antigravity).** 2. **Explicit Disclaimer:** This alpha is an empirical finding from historical data. **It is NOT canonical law, NOT an approved live EA, and NOT currently deployed.** 3. **Mandatory Next Gates Before Live Deployment Consideration:** - [ ] Forward-paper validation on FTMO MT5 demo to verify real Asian session spreads and execution latency around 00:55 UTC. - [ ] Independent audit by Claude / collaborators against FTMO overnight holding rules and drawdown gates. - [ ] Creation of a dedicated reflex executor following the Brain/Reflex architecture if approved.