Strategies

OKX ADA Volume-Confirmed Intermediate Momentum Long-Short

Outcome: Abandoned

OkxAdaVolumeConfirmedIntermediateMomentumLongShort

Outcome Summary

This strategy tried to capture intermediate-term ADA momentum confirmed by dollar-volume expansion, trading both long and short on OKX daily bars. The first backtest showed a thin but fee-clearing edge (+7.8%, Sharpe 0.20, 0.436% per trade over 299 trades), so backtest review sent it to optimization. Optimization chose the trial with the highest mark-to-market Sharpe, which had a negative return; the resulting book lost 13.9% with a 35.7% drawdown and failed the holdout (Sharpe -0.21 on 60 trades). The analyst abandoned it after one iteration as an overfit holdout failure and also flagged a suspected artifact in the engine's mark-to-market series for review.

Strategy report

Hypothesis

A single-leg ADA-USDT perpetual strategy on OKX adapting the source’s momentum-plus-turnover mechanism. Since crypto lacks point-in-time shares-outstanding data, dollar-volume expansion relative to ADA’s own trailing baseline substitutes for turnover. The strategy evaluates daily and remains flat without confirmation.

The implementation follows the stated momentum-plus-turnover mechanism using ADA dollar volume as the crypto turnover proxy. The raw 20-day return is returned continuously, volume is used only as confirmation, and the OKX perpetual is used because the hypothesis requires both long and short directions. State is bounded and updated incrementally per daily bar.

Failure code: overfit (holdout failure). By the first measured report's signature (Sharpe 0.20, PF 1.066 on 299 trades) the edge is near zero; this is not fee_edge, because avg_trade was 0.436% against the 0.15% OKX floor. The holdout Sharpe was -0.21 on 60 trades (2024-02 to 2025-02), which is a hard gate. The optimized configuration loses money over the full period (-13.9%, PF 0.94, drawdown 35.7% against the pre-registered 20%). Its 0.67 Sharpe is incoherent with its return (metrics_reliable=false). The optimizer picked the trial with the highest MTM-series Sharpe even though that trial's return was negative, so the walk-forward OOS figure of 0.61 and the DSR of 0.89 overstate the evidence. The result is not beta (beta about 0). For engine review, unverified: recurring paired -X%/+X% daily MTM swings appear in both reports, which suggests an MTM-series artifact for bracket-order strategies on OKX daily bars. The optimizer should reject trials whose Sharpe disagrees in sign with total_return. Note for the Research Lead: momentum confirmed by dollar-volume expansion on a single alt at a 20-day lookback showed no edge out of sample. A 1-3 month lookback, closer to the source, was never tested because the sweep only covered 10-30 days. If the idea is re-proposed, it should be a cross-sectional multi-coin panel at a 60-120 day lookback, not another single-ADA timing book.

Implementation

A daily ADA-USDT OKX perpetual strategy that enters long or short only when intermediate momentum exceeds a 20-day threshold and current dollar volume is at least 1.2 times its trailing 20-day baseline. It exits on momentum reversal and uses an 8% stop with a 16% take-profit bracket.

Verification Results

No code change needed. The analyst should treat per-trade economics as pre-funding. The average hold of about 2.3 days makes the impact small but not zero.

Funding is not charged in the backtest (funding_folded=false, funding_events_available=0). This is a two-sided OKX perp book with about 39% exposure, so realized PnL omits funding in both directions.

Leave the parameter tunable. The optimizer's range for momentum_lookback should include 60-250 bars.

The hypothesis says 'intermediate momentum', which in the source turnover-momentum literature means a 3-12 month horizon. The code defaults to 20 daily bars. The mechanism (a return threshold gated by the trailing dollar-volume ratio) is implemented faithfully, but the horizon is at the short end.

None required.

The exit is momentum reversal only; volume confirmation is not required to exit, which is consistent with the hypothesis. A position whose momentum stays above 0 but decays below entry_momentum is held until the sign flips, the 8% bracket stop fires or the 16% bracket take-profit fires. There is no time stop, and the hypothesis names none.

Sandbox shows PF 1.08, Sharpe 0.15 (CI -0.72 to 1.08) over 299 trades, and a 939-day drawdown. The stressed-vol tercile loses (-11.4%), and that is a soft-gate failure. Shorts win 36% of the time against 54% for longs, so the short side may be diluting a long-only trend effect. Impact is 19.5% of gross. Funding is NOT folded (funding_events_available=0 on OKX), so short and long carry over 2-day holds is unmeasured. Check per-side PnL and funding sensitivity first. The 20-day lookback is a short proxy for the 'intermediate' (months-scale) horizon in the source literature, so the optimizer should explore longer lookbacks.

Backtest Review

Sharpe: 0.20
Total return: 7.79%
Max drawdown: 11.35%
Trades: 299
Win rate: 48.2%
Profit factor: 1.07

Enough trades to optimize: 299 trades on 1794 daily bars from 2020-03 to 2025-02, about 60 trades per year. The 20% holdout should get about 60 trades, well above the floor of 10.

Per-trade return clears the OKX futures fee floor: avg_trade_return_pct is 0.436% against a floor of 0.15%, about 2.9 times. Commissions are 3.1% of gross and impact is 23% of gross, under the 50% cap. This is not a fee_edge case.

The return is not beta (L170): beta is 0.004 and benchmark correlation is 0.06. Max drawdown is 11.3%, far below ADA buy-and-hold's drawdown of more than 80% over 2021-2022. Time in market is 39%.

The trades match the hypothesis: 1-2 day momentum entries, only when dollar volume is expanding, on both sides. Funding is included in PnL (2059 funding events, and each trade record carries a funding field), so QA's note that funding is unmeasured does not apply to this run.

QA's per-side check: longs made +$7,083 over 208 trades and shorts made +$705 over 91 trades. Both sides are positive, but the shorts add very little.

The edge is thin: PF is 1.066 and total return is +7.8% over about 5 years (CAGR 1.6%). Sharpe is 0.20 with a 95% CI of -0.68 to 1.09. The stressed-volatility third of the period loses 12.6%, 2025 is at -5.7%, and the longest drawdown lasted 939 days. QA's concern is partly borne out: the edge is weak and depends on the regime. It is not falsified, because the per-trade return stays above the fee floor.

Suspected metrics artifact (cause not verified): daily_returns show regular paired swings of about -0.34%/+0.34% every 2-3 days while the book is flat. One example is 2020-08-04 to 2020-09-02, when no trade is open. The swings grow with equity, to about -3.9%/+4.1% by 2025. The equity_curve behind them is sparse. They push annualized volatility to 35.9%, far above the few percent that a 13% average position at 39% time in market would produce. This probably understates the mark-to-market Sharpe and PSR. The engine needs to review how daily_returns are built from a sparse equity curve.

The short side contributes almost nothing: 36% win rate and about 9% of PnL.

Analysis

The walk-forward is not flagged overfit: PBO is 0.07, there are no sensitivity cliffs, and the sample of 382 trades over 1,794 days is large enough.

The strategy is not beta. Beta is about 0.01 and correlation with ADA buy-and-hold is 0.06, so L170 does not apply.

The first report's per-trade return was 0.436%, which clears the OKX single-leg floor of 0.15%. The death is therefore not fee_edge.

HARD failure: the holdout Sharpe was -0.21 on 60 trades (2024-02-15 to 2025-02-07). The edge did not appear on unseen data.

The optimized book loses money: total return is -13.9%, PF is 0.936, expectancy is -$36 per trade, max drawdown is 35.7% (the pre-registered limit was 20%), and 2021, 2022, 2024 and 2025 were all negative years.

The optimizer's selection metric is incoherent. The chosen trial has the highest Sharpe of the sweep (0.673) but a negative total return (-11.5%) and PF 0.93. metrics_reliable=false says the Sharpe is computed on the mtm_equity_curve series and disagrees in sign with the equity-curve return. The walk-forward OOS figure of 0.61 and the DSR of 0.89 inherit that series, so they overstate the evidence.

Suspected MTM or ledger artifact, mechanism not verified: the daily_returns in both reports show recurring paired -X%/+X% days (for example -14.7/+17.3 in Jan 2025) during stretches where the strategy should be flat or holding a single position. avg_holding_bars is 0 while avg_holding_period is about 1.5 days. That symmetric swing inflates the MTM-series Sharpe without adding PnL. It needs engine review of the MTM equity series for bracket-order strategies on OKX daily bars.

The first report was also weak: Sharpe 0.20 (CI -0.68 to 1.09), PF 1.066 on 299 trades, and a loss in the stressed-volatility tercile (-12.6%).

The pre-registered prediction met 1 of 5 checks: OOS Sharpe, drawdown, PF and per-trade return were all missed.

The candidate does not survive programme-level FDR (it is not among the 8 kept of 342).

computed on the unreliable MTM series

> 0 on >= 10 trades

Benjamini-Hochberg at q=0.10 over 342 programme candidates keeps 8. A candidate that does not survive here is not distinguishable from the programme's own noise, however good its individual statistics look.

Outcome Summary

The optimizer should reject trials whose Sharpe disagrees in sign with total return. If the idea returns, test it as a cross-sectional multi-coin panel with a 60-120 day lookback rather than as another single-ADA timing strategy with a 10-30 day lookback.

The analyst abandoned it with failure code overfit (holdout failure): the holdout Sharpe of -0.21 fails the hard gate. The optimized book also lost money, broke the pre-registered 20% drawdown limit, did not survive programme-level FDR (8 of 342 kept), and its 0.67 Sharpe disagreed in sign with its return (metrics_reliable=false).

A daily, single-leg long-short strategy on the ADA-USDT perpetual on OKX. It traded 20-day price momentum only when dollar volume was at least 1.2x its own trailing baseline, using that expansion in place of the share-turnover measure from the source research.

The initial backtest (2020-03 to 2025-02) returned +7.8% over 299 trades, with Sharpe 0.20 (95% CI -0.68 to 1.09), PF 1.066, 0.436% average return per trade and 11.3% max drawdown. After optimization the configuration lost money: -13.9% total return, PF 0.936, 35.7% max drawdown on 382 trades, and a holdout Sharpe of -0.21 on 60 trades.

Analysis

Combining Momentum Effect with Volume

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