DOT OKX Daily SMA 5/15 — Half-Exposure Risk Replication

Hypothesis

Separately registered lower-risk replication of the existing DOT/OKX adaptation of the FMZ JavaScript state machine. Retain completed daily bars, strict SMA 5/15 crossings, signal-close market execution and exit-only opposite transitions. Change only sizing_risk_fraction from 0.01 to 0.005 and max_notional_fraction from 0.20 to 0.10. Target entry notional becomes equity × min(0.10, 0.005/0.05) × 1 = 10% of equity, versus the prior 20%. A 5% adverse instrument move therefore corresponds to approximately 0.5% of entry equity before costs; the reference move is a sizing assumption, not a stop or guaranteed loss limit. The state machine depends on the first eligible crossing: ordinary alternating crossings can repeatedly enter the initial direction and use opposite crossings only to exit. Balanced directional exposure is not claimed. This experiment tests the revised risk contract; reducing exposure does not establish an economic edge. Source figures and full text were not supplied in this revision session, so their earlier inspection claims are not independently repeated.

The task's generic summary says the premise has no edge, but its specific analyst recommendation requests a lower-risk replication with unchanged signals. Follow that narrower recommendation without inventing a mechanism change or claiming that halving exposure rescues profitability. The substantive experimental difference is the fixed allocation contract, with target entry exposure reduced by half. Retain DOT/OKX intentionally for comparability rather than claim symbol diversification; corpus concentration and portfolio distribution were not provided. Preserve all prior evidence and protected-data exposure lineage. Previously inspected holdout data cannot become unseen through a new title, allocation or registration; subsequent evaluation must retain the existing lineage and identify genuinely unexposed evidence.

Implementation

DOT/OKX completed daily-bar strict SMA 5/15 crossings. Enter long on an upward crossing while flat or short on a downward crossing while flat. An opposite crossing closes the held position without reversing on the same bar. Market orders use signal-close execution on OKX. Entry allocation is 10% of settled equity at 1x leverage; no protective stop, target, filter or time exit is added.

Implements the revised hypothesis from scratch with unchanged signal and exit rules and only the requested allocation contract: sizing_risk_fraction=0.005 and max_notional_fraction=0.10, with fixed 5/15 periods, 0.05 sizing reference and 1x leverage. A 5% adverse move implies approximately 0.5% of entry equity before costs, not a guaranteed loss limit. Alternating crossings can repeatedly enter the initial direction; balanced exposure is not claimed. Halving exposure does not establish economic edge. Prior evidence and protected-data exposure lineage must remain attached by the pipeline; this registration does not make inspected holdout data unseen. Original source text and figures were not supplied, so no independent source-fidelity claim is made.

Verification Results

Halving allocation reduces exposure but does not establish an economic edge. Full-history evaluation must retain prior protected-data exposure lineage.

Backtest Review

Sharpe

0.24

Total return

8.30%

Max drawdown

11.74%

Trades

67

Win rate

26.9%

Profit factor

1.18

67 closed trades over 1,660 days provide a functioning sample for the registered frozen evaluation. All trades are long, consistent with the explicitly direction-dependent, exit-only crossing state machine.

Entry exposure is approximately 10%, consistent with the revised sizing contract. Drawdown is 11.736%, within the declared 15% limit.

Net return is 8.295%, PF is 1.181 and average trade return is 1.429% of notional, above the supplied 0.10% round-trip cost estimate. No liquidation or terminal unrealized gain is reported.

Sharpe is only 0.241, alpha is -0.0069 and information ratio is -0.474; economic edge remains unproven. QA's warning that lower allocation does not establish an edge remains valid.

Buy-and-hold returned -7.697%; its sampled detail curve shows approximately 93.153% drawdown. Lower strategy risk alone does not demonstrate timing alpha.

Exact crossing prices and execution timing were not independently checked against source bars.

The read-only workspace prevents saving this JSON to the requested outbox path.

Analysis

The risk contract worked: max drawdown is 11.7%, under the pre-registered 15%. Peak entry leverage is 0.10 and accounting reconciles.

All four checkable pre-registered predictions were met: 67 trades, +8.3% return, 11.7% drawdown and 1.43% average trade return. The OOS Sharpe prediction cannot be checked.

Fees are not the problem. The average trade returns 1.43% against a 0.20% floor (about 0.10% round-trip cost on OKX). Commissions are 1.3% of gross and impact 3.8%.

The trades match the disclosed state machine: 67 longs, 0 shorts. The first eligible crossing was bullish and opposite crossings only exit.

Hard gate, cannot be waived: only 3 of 6 calendar years were positive (50%, below the 60% floor). 2020 and 2025 are partial years; the full years 2021-2024 were 3 of 4 positive. The gate is still classified hard in numeric_gates.

There is no measurable signal. The executed-signal IC is 0.0013 (t=0.05, n=1659), rejected as no information, with a net edge of -0.001%/bar.

There is no alpha against the benchmark: alpha is -0.69%/yr, information ratio -0.47 and beta 0.06. The return is not explained by an edge.

Profit is concentrated. 2021 carries 68% of the profit, and one month, 2024-11 (+11.2%), is larger than the whole 4.5-year net return (+8.3%). Without that month the strategy lost money.

Out-of-sample evidence is weak and inconsistent. Pooled outer OOS Sharpe is 0.12 (t=0.21, corrected p=0.45, N_oos=2). Window OOS Sharpes range from -3.6 to +2.5 on 3-5 trades each. Holdout Sharpe is 0.59 (t=0.57, p=0.28).

Profit factor is 1.18 and Calmar 0.15. The longest drawdown lasted 1379 days, and returns were negative in the normal-volatility tercile (Sharpe -0.51) and in every dated stress window except SVB.

Halving exposure from 20% to 10% is a pure scaling change. As the hypothesis itself states, it cannot create an edge, and the signal diagnostics confirm there is none to scale.

Minor reporting inconsistency, not verified: avg_holding_bars is 0.0 while avg_holding_period is 11d 13h.

soft

unevaluated in backtest metrics; treated as unknown

advisory

inconclusive

t=0.57, Sharpe 0.59

t>=2

t=0.21 vs hurdle 2.08

DSR>=Phi(2)

hard

The Research Lead REVISED this hypothesis because the previous premise did not work. Implement the REVISED hypothesis (see the hypothesis description) from scratch — do not reuse the prior strategy. Revision rationale: The task's generic summary says the premise has no edge, but its specific analyst recommendation requests a lower-risk replication with unchanged signals. Follow that narrower recommendation without inventing a mechanism change or claiming that halving exposure rescues profitability. The substantive experimental difference is the fixed allocation contract, with target entry exposure reduced by half. Retain DOT/OKX intentionally for comparability rather than claim symbol diversification; corpus concentration and portfolio distribution were not provided. Preserve all prior evidence and protected-data exposure lineage. Previously inspected holdout data cannot become unseen through a new title, allocation or registration; subsequent evaluation must retain the existing lineage and identify genuinely unexposed evidence.

Outcome Summary

Cutting exposure met the 15% drawdown contract but cannot create an edge, so another sizing-only replication would add an exposed variant without testing anything new.

The backtest review approved optimization, but the post-optimization analyst abandoned it as premise_falsified. It failed a hard gate (only 3 of 6 calendar years positive, below the 60% floor). It also showed no measurable signal (IC 0.0013, t=0.05), negative alpha against DOT buy-and-hold (-0.69%/yr, IR -0.47), and profit concentrated in November 2024, a month that alone exceeded the total net return.

A lower-risk replication of a DOT/OKX daily SMA 5/15 crossing state machine (adapted from an FMZ JavaScript strategy) that changed only sizing: entry notional was cut from 20% to 10% of equity, with opposite crossings used only to exit.

Over 1,660 days (2020-09 to 2025-03), it returned +8.3% net across 67 trades, all of them long. Sharpe was 0.24, max drawdown 11.7%, profit factor 1.18 and average trade return 1.43% of notional. Out of sample, pooled outer Sharpe was 0.12 (t=0.21) and holdout Sharpe 0.59 (t=0.57).

Analysis

手把手教你写策略--移植一个my语言策略