Strategies

Administrator SMA Oscillator Faithful Port: DOT Spot Daily Long-or-Cash

Outcome: Abandoned

DotSpotAdministratorSmaOscillator

Outcome Summary

This strategy ported Administrator's SMA(10)/SMA(21) crossover to DOT spot daily bars as a long-or-cash system with 20% equity allocation. The initial backtest looked good: 140.9% return, PF 2.21, 15.8% average trade, and much smaller drawdown than buy-and-hold. Backtest review approved it for frozen validation. Closer analysis found that the 2021 bull run supplied 94% of the profit, later years netted about -$5.0k, most OOS windows were negative, and drawdown broke the 20% pre-registered cap. With no tunable parameters and the code correctly implementing the source, the analyst abandoned it after one iteration, and risk review was never reached.

Strategy report

Hypothesis

Faithful port of Administrator's 'MACD Oscillator backtest' in je-suis-tm/quant-trading, Apache-2.0. The executable signal uses simple moving averages, not exponential averages, a MACD signal line, or a histogram crossover. The script requests ma1, ma2, dates, ticker and slicer interactively; it has no hardcoded numerical defaults. This platform experiment selects ma1=10 and ma2=21, the values recommended in the source comments, on DOTUSDT.BINANCE_SPOT daily bars. Dates use available history. The source is a tradeable long-or-cash signal generator despite plotting exits as 'SHORT'; it does not open short positions. Platform additions are equity-relative sizing and real trading costs. No stop, target, regime filter or time exit is added.

The supplied corpus_concentration is empty, so DOT's historical share cannot be established. DOT avoids adding another BTC or ETH experiment, and BINANCE_SPOT is below its 15% venue target. Single-symbol and long-only exposure remain overrepresented, but adding shorts, a basket or a cross-venue hedge would change the explicitly requested source rules. Daily bars preserve the Yahoo daily-data framing and avoid adapting this slow crossover into a fee-intensive intraday strategy. Historical failures inform these implementation choices but do not justify declining a faithful test.

premise_falsified: abandon this faithful long-or-cash port of Administrator's SMA(10)/SMA(21) oscillator on DOTUSDT.BINANCE_SPOT daily bars. The full-period figures (Sharpe 0.94, PF 2.21, avg trade 15.8% vs the 0.25% floor) come almost entirely from the 2021 DOT bull run, which holds 94% of profit and $143k of $140.9k net. The 32 trades from 2022-2025 net about -$5.0k, so they show no edge after costs. Frozen-parameter chronological OOS agrees: 6 of 9 outer windows are negative, mean OOS Sharpe is about -0.51, and the only strong window covers 2021. The procedure was rejected because the latest training window broke the pre-registered 20% drawdown cap (20.31%), and full-period drawdown of 21.99% also misses the prediction. The forecast IC has t=1.88, not distinguishable from noise. There are no tunable parameters, and the code correctly implements the source, so neither developer iteration nor re-optimization can help. Shrinking allocation to pass the drawdown cap would not create an edge that is absent after 2021. This judges only this platform operationalization (fixed 10/21 SMAs, DOT spot daily, 20% allocation). It does not judge the source author's interactive backtest script or moving-average crossovers generally.

Implementation

Daily DOT spot long-or-cash strategy using the difference between 10-row and 21-row simple moving averages. Buys on a cash-to-long transition and liquidates on a long-to-cash transition, allocating 20% of settled equity per entry.

Preserves partial-window SMA initialization, the zero-based ma1-row trading gate, the >= comparison, and differenced holding-state transitions. The continuous signal is the raw SMA oscillator. Slicer remains zero and, as in the source, does not alter signal generation. Platform adaptations are market execution, equity-relative sizing, DOT spot identifiers and engine-applied costs. No stops, targets, filters, shorts or time exits are added; the research plan remains frozen.

Backtest Review

Sharpe: 0.94
Total return: 140.91%
Max drawdown: 21.99%
Trades: 42
Win rate: 33.3%
Profit factor: 2.21

42 closed longs over 1659 days are reasonable for daily SMA transitions. Spot-checked entries and exits match oscillator sign changes; quantities reflect 20% equity-relative allocation.

Net average trade return is 15.80% versus the supplied 0.25% spot review floor; PF is 2.21 and Sharpe is 0.94. Closed-trade PnL reconciles to the 140.91% return, with no ending unrealized gain.

Buy-and-hold returned 7.97% with approximately 93.15% drawdown on the saved curve, versus strategy drawdown of 21.99%.

One 92-day bull-market trade contributes 87.3% of net profit. Forward validation must establish whether performance survives this concentration.

Drawdown exceeds the declared 20% risk limit; the one-day forecast diagnostic is inconclusive at t=1.88.

The registered plan has no tunable parameters: proceed with frozen chronological validation, preserving SMA 10/21 and allocation 0.20.

The verdict could not be saved to the requested outbox because the filesystem is read-only.

Analysis

The port is faithful. It has 42 long-only closed positions over 1,659 days, entries and exits fall on SMA(10)/SMA(21) sign changes, sizing is 20% of equity, and there are no shorts.

Accounting is valid: reconciles, end_unrealized_pct is 0, commissions are 1.45% of gross and modeled impact is 0.91%. Full-period avg_trade_return_pct is 15.8%, against the 0.25% spot floor and 0.7% predicted. PF is 2.21, and the payoff ratio of 4.4 carries a 33% win rate.

One year carries the whole result. 2021 contributes 94% of profit (+$143.3k realized over 7 trades). Over 2022-2025, the 32 closed trades net about -$5.0k: 2022 -$33.1k, 2023 +$5.2k, 2024 +$29.8k, 2025 -$6.9k. Excluding the 2021 DOT bull run, the crossover has no measurable edge.

Chronological frozen-parameter OOS confirms this. Nine outer windows have OOS Sharpe 2.41, -2.57, -0.50, -0.55, -2.19, 0.57, -1.56, -1.91 and 1.67 (mean about -0.51, 6 of 9 negative). The only strong positive window covers 2021.

The optimization procedure was rejected as insufficient_candidate_coverage. The latest training window failed the pre-registered 20% drawdown constraint (inner drawdown 20.31%), so there is no latest candidate and no holdout was run. numeric_gates has a hard failure (no valid optimization report), so promotion is impossible.

Full-period max drawdown is 21.99%, above the pre-registered 20% (the prediction missed). The drawdown lasted 1,236 days without recovery.

The signal-research diagnostic shows IC 0.046 (t=1.88), not distinguishable from noise. Information ratio vs buy-and-hold is -0.345. Sharpe 0.94 has a 95% CI of 0.10 to 1.82.

insufficient_candidate_coverage

valid

Outcome Summary

A strong full-period result from one bull-market year (here, DOT in 2021) is not an edge, so check year-by-year PnL concentration and frozen chronological OOS before trusting aggregate Sharpe or per-trade returns.

Backtest review approved it for frozen-parameter validation, but the analyst then abandoned it as premise_falsified. 2021 produced 94% of the profit, 6 of 9 frozen-parameter OOS windows were negative (mean Sharpe about -0.51), and the optimization procedure was rejected as insufficient_candidate_coverage because the latest training window broke the pre-registered 20% drawdown cap (20.31%), so no holdout was run.

A faithful long-or-cash port of Administrator's SMA(10)/SMA(21) oscillator from je-suis-tm/quant-trading on DOTUSDT.BINANCE_SPOT daily bars. It held DOT while the fast SMA was at or above the slow SMA, sized at 20% of equity, with no stop, target or regime filter.

Over 1,659 days (2020-09 to 2025-03) it returned 140.9% net with Sharpe 0.94 (95% CI 0.10-1.82), profit factor 2.21, 42 long trades, a 33% win rate and an average trade return of 15.8% of notional. Max drawdown was 21.99%, and the 2022-2025 trades netted about -$5.0k.

Analysis

MACD Oscillator backtest

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