ETH OKX vn.py AtrRsiStrategy — Frozen Source-Default Translation

Faithful signal and order translation of AtrRsiStrategy by 用Python的交易员, MIT licensed. Source: https://github.com/vnpy/vnpy_ctastrategy/blob/6ef76981624bf55b2ea978f8587f74d633aafc72/vnpy_ctastrategy/strategies/atr_rsi_strategy.py. Preserve atr_length=22, atr_ma_length=10, rsi_length=5, rsi_entry=16, trailing_percent=0.8 and fixed_size=1. Use ETHUSDT.OKX perpetual with completed 1-MINUTE bars, matching the source's default tick-to-minute BarGenerator. No higher-timeframe aggregation, additional filters or parameter fitting. Platform sizing is the explicitly authorized adaptation: fixed_size=1 represents one capital-relative allocation, sized at entry to 10% of current equity in notional, rounded down to instrument quantity precision. Hold that quantity until exit; never pyramid. Config leverage=1.0. Starting equity is $100,000. Source initialization requests ten days of historical bars; maintain a rolling 100-bar ArrayManager-equivalent buffer and prohibit trading until it is initialized. Logging and event callbacks do not generate signals; order and stop-order callbacks are otherwise no-ops.

verification_loop: Verification failed (Layer 4 — QA review) [class=hypothesis_mismatch]: - [edge_concern] The five-day smoke test returned -0.139% per trade over 38 closed positions, with PF 0.626. Full-history evaluation should examine turnover and trailing-stop giveback; these results do not independently block the strategy. - [critical] The crossed-stop fallback does not preserve source execution. _submit_exit() immediately calls close_position(); OKX has zero configured latency, so Nautilus executes against the current market at the decision bar's close, rather than the next bar's open. Furthermore, vn.py checks the renewed stop against the next bar's range: a gap back above a long stop can leave it untriggered. An unconditional market exit changes both timing and trigger conditions. Source: https://raw.githubusercontent.com/vnpy/vnpy_ctastrategy/6ef76981624bf55b2ea978f8587f74d633aafc72/vnpy_ctastrategy/backtesting.py (line 163) — fix: Implement pending source-style stops evaluated against subsequent market data, preserving conditional triggering and gap-price execution. Add an engine execution hook if necessary; do not substitute an immediate market close. Verify ordinary crossing, gap-through, and gap-away cases for both directions. - [critical] BUY limits at close+$5 and SELL limits at close-$5 are marketable when submitted to Nautilus's current close-price book. With post_only=False and zero OKX latency, they can fill on the signal bar. The source backtester matches existing orders before invoking on_bar, so newly generated limits first become eligible on the subsequent bar, with their price constraint still enforced. This changes entries, gap handling, and subsequent trailing-stop state. Source: https://raw.githubusercontent.com/vnpy/vnpy_ctastrategy/6ef76981624bf55b2ea978f8587f74d633aafc72/vnpy_ctastrategy/backtesting.py (line 144) — fix: Provide subsequent-bar activation for the original limit price and entry-sized quantity. Validate that no entry fills on the signal bar and that a next-bar gap beyond the limit prevents an opening fill unless the subsequent range reaches the limit.

Verification failed (Layer 4 — QA review) [class=hypothesis_mismatch]: - [edge_concern] The five-day smoke test returned -0.139% per trade over 38 closed positions, with PF 0.626. Full-history evaluation should examine turnover and trailing-stop giveback; these results do not independently block the strategy. - [critical] The crossed-stop fallback does not preserve source execution. _submit_exit() immediately calls close_position(); OKX has zero configured latency, so Nautilus executes against the current market at the decision bar's close, rather than the next bar's open. Furthermore, vn.py checks the renewed stop against the next bar's range: a gap back above a long stop can leave it untriggered. An unconditional market exit changes both timing and trigger conditions. Source: https://raw.githubusercontent.com/vnpy/vnpy_ctastrategy/6ef76981624bf55b2ea978f8587f74d633aafc72/vnpy_ctastrategy/backtesting.py (line 163) — fix: Implement pending source-style stops evaluated against subsequent market data, preserving conditional triggering and gap-price execution. Add an engine execution hook if necessary; do not substitute an immediate market close. Verify ordinary crossing, gap-through, and gap-away cases for both directions. - [critical] BUY limits at close+$5 and SELL limits at close-$5 are marketable when submitted to Nautilus's current close-price book. With post_only=False and zero OKX latency, they can fill on the signal bar. The source backtester matches existing orders before invoking on_bar, so newly generated limits first become eligible on the subsequent bar, with their price constraint still enforced. This changes entries, gap handling, and subsequent trailing-stop state. Source: https://raw.githubusercontent.com/vnpy/vnpy_ctastrategy/6ef76981624bf55b2ea978f8587f74d633aafc72/vnpy_ctastrategy/backtesting.py (line 144) — fix: Provide subsequent-bar activation for the original limit price and entry-sized quantity. Validate that no entry fills on the signal bar and that a next-bar gap beyond the limit prevents an opening fill unless the subsequent range reaches the limit.

Verification failed (Layer 4 — QA review) [class=code_defect]: - [edge_concern] Sandbox: 38 trades in 5 days, avg_trade_return_pct -0.139%, PF 0.63, commissions 43% of gross. Both entry (a limit at close±$5, so it crosses and fills as taker) and exit (stop-market) pay taker fees, about 0.10% round trip. Each trade gives back up to 0.8% through the trailing stop, and a 1-minute RSI(5)>66 signal has to clear that cost about 7 times a day. Expect thousands of trades over the ~2,444-day history. After the stop fix, check gross per-trade return against the 0.15% floor first. With trade counts this high, a PF near 1 would be a falsified measurement, not something to tune. - [critical] The trailing stop is resubmitted on every bar as a standalone STOP_MARKET at intra_trade_high*(1-0.8%) for longs (or intra_trade_low*(1+0.8%) for shorts). That level includes the current bar's high (or low). If the bar closes more than 0.8% below its own high, the new long stop is already in the market when it is submitted. The OKX backtest venue runs with NautilusTrader's default reject_stop_orders=True; in engine.pyx _process_stop_market_order, a triggered stop is REJECTED, not filled. should_exit() always returns False, so the position has no exit that bar. On later bars the stop level can only rise (longs) or fall (shorts), so it stays in the market and keeps getting rejected until price recovers above it. The position can bleed without limit. In vn.py, the same stop triggers on the next bar and fills at about the open. The hypothesis's only exit, the 0.8% trailing stop, is therefore not implemented reliably. This hits exactly the volatile ATR>ATR_MA regime the strategy selects, including a signal bar with a long wick being the first stop bar. The 5-day sandbox did not happen to hit it (largest loss -0.96% of notional), but over ~2,444 days of 1-minute ETH it will. (line 186) — fix: Before calling _submit_stop_market, test whether the stop is already crossed. For a long, check stop >= current_close; for a short, check stop <= current_close. If it is, call self._submit_exit(position) (a reduce-only market close), which matches vn.py's next-bar stop trigger. Otherwise submit the resting stop as you do now. As a safety net, also track rejected stop orders, e.g. in on_order_rejected, and flatten the position. - [warning] entry_price_offset is a fixed $5. That is about 0.13-0.33% of ETH's price across its history, so the entry limit always crosses and fills as taker at the next bar. This is faithful to the source and is not a defect. Note that vn.py fills at min(limit, next open), and Nautilus behaves similarly. (line 133) — fix: No change needed; recorded for the analyst.

The trailing stop is resubmitted on every bar as a standalone STOP_MARKET at intra_trade_high*(1-0.8%) for longs (or intra_trade_low*(1+0.8%) for shorts). That level includes the current bar's high (or low). If the bar closes more than 0.8% below its own high, the new long stop is already in the market when it is submitted. The OKX backtest venue runs with NautilusTrader's default reject_stop_orders=True; in engine.pyx _process_stop_market_order, a triggered stop is REJECTED, not filled. should_exit() always returns False, so the position has no exit that bar. On later bars the stop level can only rise (longs) or fall (shorts), so it stays in the market and keeps getting rejected until price recovers above it. The position can bleed without limit. In vn.py, the same stop triggers on the next bar and fills at about the open. The hypothesis's only exit, the 0.8% trailing stop, is therefore not implemented reliably. This hits exactly the volatile ATR>ATR_MA regime the strategy selects, including a signal bar with a long wick being the first stop bar. The 5-day sandbox did not happen to hit it (largest loss -0.96% of notional), but over ~2,444 days of 1-minute ETH it will.

verification_loop: Verification failed (Layer 4 — QA review) [class=hypothesis_mismatch]: - [edge_concern] The five-day smoke test returned -0.139% per trade over 38 closed positions, with PF 0.626. Full-history evaluation should examine turnover and trailing-stop giveback; these results do not independently block the strategy. - [critical] The crossed-stop fallback does not preserve source execution. _submit_exit() immediately calls close_position(); OKX has zero configured latency, so Nautilus executes against the current market at the decision bar's close, rather than the next bar's open. Furthermore, vn.py checks the renewed stop against the next bar's range: a gap back above a long stop can leave it untriggered. An unconditional market exit changes both timing and trigger conditions. Source: https://raw.githubusercontent.com/vnpy/vnpy_ctastrategy/6ef76981624bf55b2ea978f8587f74d633aafc72/vnpy_ctastrategy/backtesting.py (line 163) — fix: Implement pending source-style stops evaluated against subsequent market data, preserving conditional triggering and gap-price execution. Add an engine execution hook if necessary; do not substitute an immediate market close. Verify ordinary crossing, gap-through, and gap-away cases for both directions. - [critical] BUY limits at close+$5 and SELL limits at close-$5 are marketable when submitted to Nautilus's current close-price book. With post_only=False and zero OKX latency, they can fill on the signal bar. The source backtester matches existing orders before invoking on_bar, so newly generated limits first become eligible on the subsequent bar, with their price constraint still enforced. This changes entries, gap handling, and subsequent trailing-stop state. Source: https://raw.githubusercontent.com/vnpy/vnpy_ctastrategy/6ef76981624bf55b2ea978f8587f74d633aafc72/vnpy_ctastrategy/backtesting.py (line 144) — fix: Provide subsequent-bar activation for the original limit price and entry-sized quantity. Validate that no entry fills on the signal bar and that a next-bar gap beyond the limit prevents an opening fill unless the subsequent range reaches the limit.