Strategies

NEAR OKX MultiTimeframeStrategy Faithful 5m RSI / 15m SMA Port

Outcome: Abandoned

NearOkxMultiTimeframeFaithful

Outcome Summary

NearOkxMultiTimeframeFaithful preserved the source strategy's 5-minute RSI and 15-minute SMA rules while adapting the instrument and position sizing. After one iteration, its backtest recorded 7,299 closed trades, a -98.989% return, Sharpe -10.354 and 99.013% drawdown. Backtest review abandoned it for negative expectancy and the absence of a justified optimization path, so later stages were not reached. The rejection applies to this specified adaptation, without asserting a translation defect or rejecting the original strategy on an unspecified market.

Strategy report

Hypothesis

Faithful port of MultiTimeframeStrategy by 用Python的交易员, MIT licensed, from https://github.com/51bitquant/howtrader/blob/fcad8e2d86825506b9d33441ab78be853c2c3bdd/examples/strategies/multi_timeframe_strategy.py. Preserve the 5-minute execution clock, 15-minute trend clock, RSI thresholds, SMA defaults, aggressive limit orders and signal-based exits. Adapt only the unspecified market to NEARUSDT.OKX and the original fixed_size=1 quantity to capital-relative sizing. Aggregate completed 1-minute bars into both clocks, invoking the 5-minute callback before the 15-minute callback at shared boundaries, exactly as the source's on_bar ordering does. Consequently, the shared-boundary trading decision uses the previously computed 15-minute trend. Preserve ArrayManager's default 100-bar initialization for each timeframe and its TA-Lib RSI/SMA semantics; verify helper defaults against the pinned source during translation. The source requests 10 days of initialization history. No supplementary signal, stop-loss, target, trailing stop, maximum holding period or reversal-on-the-same-callback is added. Performance is unverified.

The seed specifies exact rules, so preserve its mechanism and defaults despite adverse results from related RSI momentum strategies. OKX is underrepresented at 2.6%, and long-short proposals address the portfolio's 80.8% long-only concentration. NEAR avoids the repeatedly used BTC, ETH, LINK and SOL settings without claiming proven instrument-level novelty: corpus_concentration is empty. Hyperliquid's sub-hourly history limit makes it unsuitable for this faithful 5m/15m port. A single instrument preserves the source rather than inventing a basket or cross-venue hedge.

Negative expectancy on this NEAR OKX instrument-and-sizing adaptation: 7,299 closed trades average -0.627% net versus the +0.15% futures review floor, with PF 0.327. Gross P&L remains negative after recorded costs are added back, so execution costs alone do not explain the failure. The fixed-rule contract offers no justified optimization path. This rejects the specified adaptation, not the original author’s strategy on an unspecified market. No accounting or translation defect is asserted.

Implementation

NEAR OKX long-short translation using five-minute RSI(14) thresholds of 70/30 and fifteen-minute SMA(5/20) trend. Cancels orders on each execution callback, submits aggressive GTC limits, and exits on opposing trend or RSI crossing 50. Entries allocate 10% of settled equity. No additional risk or signal filters.

Preserves the supplied source's rules, bearish SMA equality, strict exit comparisons, cancellation sequence and five-minute-before-fifteen-minute callback ordering. Calendar-aligned one-minute closes provide the aggregate closes consumed by both indicators. RSI uses bounded 100-close windows with Wilder initialization and smoothing; SMA uses the trailing arithmetic mean. Indicator windows and clock state are stored as scalars for restart persistence. The ten-day initialization request is represented by processing the first ten calendar days without trading. fixed_size=1 remains provenance metadata; executable sizing uses the required capital-relative adaptation. Performance is unverified. The pinned helper URL could not be retrieved, so its default and TA-Lib-wrapper semantics were not independently verified. The read-only filesystem prevented writing the requested outbox file; the complete bound JSON is returned here.

Verification Results

The truncated sandbox produced 13 closed trades averaging -0.536% of notional, with PF 0.372. Modeled impact materially contributed to losses; investigate execution costs and sizing in the full-history review. These results do not establish a correctness defect.

Backtest Review

Sharpe: -10.35
Total return: -98.99%
Max drawdown: 99.01%
Trades: 7299
Win rate: 22.0%
Profit factor: 0.33

Reliable full-history sample: 7,299 closed trades, reconciled accounting and no terminal open position.

Trade records show both directions, five-minute entry timing and approximately 10% capital-relative sizing.

Average net trade return -0.627% versus the +0.15% futures review floor; PF 0.327, Sharpe -10.354 and total return -98.989%.

Both directions and every reported calendar year lose. QA’s economic concern is confirmed.

Adding back impact, commissions and net funding expense still leaves approximately $20,171 of gross loss.

Drawdown is 99.013% versus the registered 15% limit. Sampled buy-and-hold return is +292.66%, with approximately 94.94% drawdown.

The contract has no tunables. No implementation defect is established; the pinned upstream source could not be independently retrieved during this review.

Outcome Summary

This adaptation lost money even before recorded costs, so execution costs alone do not explain its failure.

Backtest review abandoned it because expectancy was negative, drawdown exceeded the registered 15% limit, and the fixed-rule contract offered no justified optimization path. Optimization, post-optimization analysis and risk review were not reached; no accounting or translation defect was asserted.

It adapted a strategy combining 5-minute RSI signals with a 15-minute SMA trend filter to NEARUSDT.OKX, preserving aggressive limit orders and signal exits while using capital-relative sizing.

Across 7,299 closed trades, the backtest returned -98.989% with Sharpe -10.354 and maximum drawdown 99.013%. Average net trade return was -0.627% and profit factor was 0.327; gross P&L remained negative after recorded costs were added back.

Analysis

MultiTimeframeStrategy

Backtest and paper results are hypothetical. Trading involves risk of loss.