LinkDailyTrendFollowVolExpansionLong

LINK Weekly Trend Following with Volatility Expansion Filter

A simple long-only trend-following strategy on LINKUSDT perpetual futures using daily (1D) bars. LINK (Chainlink) has well-documented trend persistence during oracle/DeFi narrative cycles — when LINK breaks out of accumulation, it tends to run for weeks. The strategy enters when price is above its 50-day SMA AND the 14-day ATR is meaningfully expanded versus the 60-day median ATR (volatility expansion regime, where trend-following has positive expectancy). It exits when price closes below the 20-day SMA OR when volatility contracts back to baseline (regime change). This uses only OHLCV bars (no supplementary data dependency, eliminating verification-timeout risk), has only 3 core parameters (50/20 MA periods + ATR ratio threshold), uses daily bars (multi-day holding period makes the ~0.10% round-trip fee negligible against typical 5-15% trend captures), and is long-only/single-instrument (simplest possible execution path). Differs from existing portfolio: not a pairs trade (unlike ADA-DOT, ETH-BTC), uses a fresh symbol (LINK), uses a regime filter rather than naive Donchian breakout (unlike NEAR), and trend-following adds diversification to mean-reversion-heavy successes.

Implements LINK daily trend-following but conditions entries on a volatility-expansion regime, where breakouts/trends actually persist, rather than a naive Donchian breakout (distinct from the NEAR Donchian strategy). The 14-ATR vs 60-day-median-ATR ratio is the core regime filter; the 50-SMA confirms uptrend. The symmetric exits treat regime change as a first-class exit: a 20-SMA close-below for trend break and a vol-contraction exit when ATR falls back to baseline, plus a catastrophic ATR stop and time stop for safety. calculate_signal returns the continuous ATR-normalized trend-strength score so the verifier sees a real, bar-varying signal, while the discrete trend+expansion boolean gates should_enter and position_size, avoiding the gated-to-0.0 anti-pattern; a rolling deque of per-bar ATR supplies the median-ATR baseline. Venue LINKUSDT.BINANCE USD-M daily is a clean long-listed perp; futures RT taker (~0.10%) is negligible against multi-day 5-15% trend captures and cheaper than spot; long-only with leverage left at 1.0 (no amplification) so the leverage_set_but_unused gate does not trigger. Only ~3 core parameters, single instrument/timeframe, OHLCV-only -- avoiding supplementary-data verification-timeout risk and cross-asset audit_stale risk, and adding trend-following diversification to a mean-reversion-heavy book.

Failed deflated Sharpe: DSR=0.0 (vs 0.95 bar), expected-max=5.49 over 225 trials — the optimized Sharpe of 1.52 is far below the best-of-N luck bar, indistinguishable from selection noise (is_significant=false). Every forward gate confirms the front-loading flagged at pre-optimization: PBO 0.5264 (>0.5), sharpe_ci_low -0.3156 (CI straddles 0), walk-forward is_overfitted=TRUE with avg IS 2.34 collapsing to avg OOS -0.45 (windows [-4.98, 1.21, 2.42], first catastrophic), and a HOLDOUT that FAILED with a catastrophically negative Sharpe -6.94. Sensitivity passed overall but has a cliff on the CORE vol-expansion parameter expand_ratio (0.34 -> 0.73 across the grid). Root cause is fat-tail outlier dependence: the optimizer thinned the book to 34 trades whose return rests on ~3 days (2024-12-18 +4.42%, 2020-08-21 +3.93%, 2023-11-17 +2.97%; return_kurtosis 59.7), with 2025 already negative (-0.89%) — DSR's skew/kurtosis adjustment correctly discounts this to 0 and the -6.94 holdout shows it does not survive forward. Not promote: DSR below the luck bar + PBO>0.5 + CI straddling 0 + is_overfitted with negative average OOS + a deeply negative holdout are each disqualifying. Not iterate (attempt 1 of 2): the failure is intrinsic fat-tail fragility + a catastrophic holdout + a core-param (expand_ratio) cliff, not a robust region the sweep under-explored — with OOS averaging negative (one window -4.98) and the edge resting on ~3 outlier days, a second best-of-225 sweep re-selects another thin, outlier-carried, sub-luck-bar config that fails forward. Not revise_hypothesis: this is a thin single-asset long-only trend-following edge with no proven sibling stranded on a dead target — it simply does not deflate, and reframing onto another alt would re-fit the same fat tails. FAILURE PATTERN: long-only daily trend-following with an ATR vol-expansion filter on a high-beta alt (LINK) shows an encouraging non-overfit BASE (Sharpe 1.81, positive CI lower bound) but, after best-of-225 selection, thins to ~34 trades whose Sharpe is carried by a handful of explosive trend days (kurtosis 60) — so it fails deflated Sharpe (0.0, far below a 5.49 luck bar), PBO 0.53, walk-forward is_overfitted with negative average OOS, and a catastrophic holdout (-6.94), with a cliff on the core vol-expansion parameter. A clean base and a smooth sensitivity grid do not rescue a fat-tail trend edge; this is the same outlier-luck signature as the abandoned NEAR Supertrend (kurtosis 175) and SOL Keltner (kurtosis 124) trend siblings, in a more decisively holdout-failing form.

Long-only daily trend-following with a volatility-expansion regime filter on LINKUSDT.BINANCE USD-M futures, pure OHLCV single feed. Enters LONG when close > 50-day SMA AND the 14-day ATR is at least 1.1x its 60-day median (vol-expansion regime where trend-following has positive expectancy). Exits when close < 20-day SMA (trend break), the ATR contracts to <= 0.9x its median (regime change), a 2.5x ATR catastrophic stop, or a 120-bar time stop. The signal is the continuous ATR-normalized trend strength (close - 50SMA)/ATR recomputed every bar; the trend+vol-expansion condition is the discrete entry gate. Sized at 20% equity notional, capped so ATR-stop risk <= 1.5% of equity. leverage 1.0.

Genuine positive edge at base: Sharpe 0.51, profit_factor 1.42, positive expectancy (+$128/trade), max drawdown only 10.2%, Sortino 1.70.

Adequate non-sparse sample: 91 trades over 6.4 years; trades match the hypothesis (long-only vol-expansion-gated trend entries); code correct (no gated-to-0.0 anti-pattern).

Returns distributed across years (positive 2020/2023/2024/2025) rather than front-loaded into a single year/month — no obvious one-shot artifact.

Favorable trend payoff geometry (avg_win 2.4x avg_loss at 37% win rate); conservative sizing (avg_position_pct 7.3%) leaves room to scale; diversifies the mean-reversion-heavy book.

Edge is somewhat outlier-dependent: return_kurtosis 105, skew 9.2, tail_ratio 0.31 — a few big days (2024-12-19 +8.3%, 2023-11-21 +6.3%) carry much of the return.

max_consecutive_losses 12 and sharpe_ci_low -0.99 (CI straddles 0) — modest standalone Sharpe; will need to clear deflated-Sharpe/walk-forward/holdout to promote.

Some negative years (2021 -4.2%, 2022 -2.7%, 2026 -2.4%) — edge is real but modest.

Base strategy had a genuine positive edge (Sharpe 1.81, profit_factor 1.48, positive expectancy, sharpe_ci_low > 0), which legitimately earned the optimization slot.

Sensitivity passed overall (cliff_count 1) and optimized risk metrics look clean in-sample (max_DD 3.4%, Sortino 7.54, profit_factor 2.94); code correct and trades match the mechanism.

Deflated Sharpe = 0.0 (is_significant FALSE); optimized Sharpe 1.52 far below the expected-max luck bar of 5.49 over 225 trials.

PBO = 0.5264 (> 0.5 — more likely overfit than not).

Walk-forward is_overfitted = TRUE: avg IS 2.34 -> avg OOS -0.45, with OOS windows [-4.98, 1.21, 2.42] (first window catastrophic).

Holdout FAILED with a catastrophically negative Sharpe -6.94 (ratio 0).

sharpe_ci_low = -0.3156 (CI straddles 0).

Fat-tail outlier dependence: optimizer thinned to 34 trades carried by ~3 days (2024-12-18 +4.42%, 2020-08-21 +3.93%, 2023-11-17 +2.97%), return_kurtosis 59.7; 2025 negative (-0.89%).

The one sensitivity cliff is on expand_ratio — the CORE vol-expansion regime parameter (0.34 -> 0.73 across the grid).

LinkDailyTrendFollowVolExpansionLong was a deliberate diversifier — a simple, three-parameter long-only trend follower gated by an ATR volatility-expansion filter, added to balance a mean-reversion-heavy book. It was the strongest candidate in this batch at the base level, with a real positive edge (Sharpe 1.81, positive expectancy, returns spread across multiple years) that legitimately earned its optimization slot. But optimization exposed intrinsic fat-tail fragility: the book thinned to 34 trades whose performance hinged on roughly three big trend days (kurtosis ~60), and it failed deflated Sharpe (0.0 vs a 5.49 luck bar), PBO 0.53, walk-forward (IS 2.34 → OOS -0.45), and most decisively a holdout that printed -6.94 Sharpe. The analyst abandoned it as the same outlier-luck signature seen in the abandoned NEAR Supertrend and SOL Keltner trend siblings — a thin trend edge that simply does not survive forward.

A clean, non-overfit base backtest and a smooth sensitivity grid do not rescue a fat-tail trend edge: when a strategy's return rests on a handful of explosive outlier days, best-of-N optimization re-selects a thin, outlier-carried config that DSR discounts to zero and that collapses in walk-forward and holdout.

It passed the backtest-review gate (verdict: optimize) but the analyst abandoned it after optimization for failing every forward gate: deflated Sharpe 0.0 vs a 5.49 expected-max luck bar over 225 trials, PBO 0.53, sharpe_ci_low -0.32, walk-forward is_overfitted=true (avg OOS -0.45 with a -4.98 first window), a catastrophically failed holdout (Sharpe -6.94), and a sensitivity cliff on the core vol-expansion parameter (expand_ratio).

A long-only LINKUSDT daily trend-following strategy that entered only in a volatility-expansion regime — price above its 50-day SMA AND 14-day ATR ≥ 1.1× its 60-day median ATR — and exited on a close below the 20-day SMA, ATR contraction back to baseline, a catastrophic ATR stop, or a time stop, using just three core parameters.

The base backtest showed a genuine edge over ~6.4 years: 91 trades, Sharpe 0.51 (reported 1.81 at the optimization stage), profit factor 1.42, positive expectancy (+$128/trade), a favorable 2.4:1 win/loss ratio at a 37% win rate, and only 10.2% max drawdown. Optimization thinned it to 34 trades with clean in-sample metrics (Sharpe 1.52, profit factor 2.94, max drawdown 3.4%), but the return rested on ~3 explosive trend days (return kurtosis ~60).