VolatilityManagedIwmExposureOverlayUseq

Volatility-Managed IWM Exposure Overlay (USEQ, Daily Bars) — Continuous Inverse-Realized-Vol Position Scaling to a Self-Calibrated Vol Target, Cash Sleeve When Scaled Down, Commission-Free, Low-Parameter

A single-instrument, long/flat exposure-scaling overlay on IWM.USEQ (Russell 2000 ETF) implementing the Moreira & Muir (2017, Journal of Finance) volatility-managed portfolio anomaly. This is NOT asset rotation and NOT regime hysteresis (the SPY/QQQ->TLT/GLD rotation strategies already in the pipeline are a different mechanism); it is a CONTINUOUS scaling of a single equity's notional by the inverse of its own trailing realized volatility toward a self-calibrated target vol. Deliberately chosen to fill the two furthest-from-target buckets that also structurally dodge the factory's two leading killers: (1) VENUE — USEQ is 1.4% of the corpus vs a BINANCE that is 70.2% (over the 40% cap), and USEQ is commission-free, which structurally sidesteps fee_edge (485 deaths, the #1 killer that overwhelmingly kills fast crypto mechanisms whose raw move is smaller than fees); (2) OVERFIT (385 deaths, #2 killer) — this is a documented academic anomaly with near-zero free parameters (the vol target is self-calibrated from the asset's own long-run realized vol, not tuned), robust across lookbacks/assets/sub-periods and out-of-sample in the literature, on DECADES of daily IWM history. IWM is chosen over SPY/QQQ because small-caps have higher and more strongly clustered realized vol, giving vol-management more to work with, and because it is a fresh corpus instrument (SPY/QQQ already used by pipeline rotation strategies).

Direct implementation of Moreira & Muir (2017) inverse-variance scaling on a single instrument, exactly as the hypothesis frames it: continuous notional scaling toward a self-calibrated vol target, NOT rotation and NOT regime hysteresis (the only discrete element is the cash floor, which is what 'cash sleeve when scaled down' requires, plus a one-band re-entry hysteresis so the floor itself does not churn). vol_exponent defaults to 2.0 = the paper's inverse-VARIANCE form; the target vol is never a tuned constant -- it is the expanding mean of IWM's own realized-vol series, read strictly point-in-time, so it carries no look-ahead and no asset-specific calibration. All state is incremental (deque of squared returns + running sums, O(1) per bar) so the 300s sandbox limit is not at risk, and the whole model is six declared parameters, each with its clamp mirrored in _param_bounds. Venue: USEQ rather than the hypothesis's implied equity default is unchanged -- IWM is a US ETF, long-only and unlevered, and USEQ's commission-free / whole-share / session-calendar semantics are what the mechanism needs. One honest caveat the Analyst should weigh: on a replay of the exact mechanism over IWM's full 2000-2026 daily history the vol-managed sleeve roughly MATCHES buy-and-hold Sharpe (~0.45 vs ~0.42) while cutting max drawdown from ~60% to ~40% and giving up ~1%/yr of CAGR to the uninvested cash sleeve -- i.e. the anomaly shows up here as risk reduction rather than as a Sharpe improvement that would clear a 0.5 OOS floor. I implemented what was asked; the call on whether that risk/return trade is promotable is the Analyst's. Trade counts (closed round trips) are ~36 over the full history and ~5 in the 2400-day sandbox window, so the >=1-trade sandbox gate and the 30-trade daily sample floor are both met.

Overfit with no active edge above market beta. Four hard validity gates fail (OOS Sharpe 0.309 < 0.5, walk-forward is_overfitted=true IS 0.77->OOS 0.309, PBO 0.60 > 0.5, holdout 9 trades < 10), so promotion is impossible. More fundamentally the strategy has NO alpha: information_ratio -0.17, alpha 0.012, beta 0.49, benchmark_correlation 0.73 -- it is a scaled-down long IWM sleeve that underperforms simply holding IWM on a risk-adjusted basis, and it fails deflation (DSR 0.454, is_significant=false) and programme-level FDR. The vol-management overlay dampens drawdown but does not manufacture the Sharpe premium the anomaly promises on this single instrument -- the entire sensitivity grid is uniformly mediocre (0.27-0.34) with no robust high-edge region, so there is nothing for the developer to tune toward (iterate cannot lift an edge that is absent across the whole parameter space). Not revise_hypothesis: the same volatility-managed-portfolio mechanism has already been tried and abandoned on ETH spot (risk_reject), QQQ (fee_edge) and a 6-major basket (risk_reject x2); reframing onto yet another instrument re-produces scaled beta rather than alpha. The underpowered-holdout provenance note (forward paper as the 'real OOS') would only rescue an otherwise-significant strategy -- this one is not distinguishable from noise or from scaled buy-and-hold.

Volatility-managed long/flat exposure overlay on IWM (Russell 2000 ETF) on daily USEQ session bars. Each session the strategy measures IWM's own trailing realized volatility (zero-mean sigma of the last 21 daily log returns, annualized) and compares it to a SELF-CALIBRATED target: the expanding, point-in-time mean of that same realized-vol series over every session observed so far (a 252-session warm-up before the target is trusted). The target sleeve weight is (target_vol / realized_vol) ** vol_exponent, capped at max_weight (0.95, a structural cash buffer -- the book is never levered, config leverage = 1.0). Flat, it buys that weight of mark-to-market equity in whole shares once the weight clears min_hold_weight + rebalance_band (re-entry hysteresis). Invested, it RESCALES IN PLACE by trading only the delta between held and target shares, and only when that delta exceeds rebalance_band of equity -- so continuous scaling costs spread/impact on the delta rather than a full round trip. When realized vol runs so hot that the target weight drops below min_hold_weight (0.65 -> realized vol ~1.24x its long-run mean at exponent 2.0) the sleeve goes fully to cash (no cash yield modelled). USEQ is commission-free, so the only trading cost is spread + impact.

Verify the base template's use of min_bars_required is limited to gating on_bar evaluation; if so this is fine, otherwise rename to an internal helper.

min_bars_required(self) overrides the framework method (static-analysis base_shadow warning). The override deliberately returns only vol_lookback+2 so the engine begins evaluating early while the much longer calib_min_sessions warm-up is gated internally by emitting a 0.0 weight. Behaviorally sound (no entries fire until calibration completes) but it is a base-method shadow — confirm the base does not rely on its own implementation for anything beyond the warm-up bar count.

No fix required; consider documenting that trade-count and holding-period metrics are not meaningful for an in-place-rescaling overlay.

should_exit() has a side effect: when still invested it calls _rescale_to_weight() (which submits BUY/SELL delta orders) before returning False. This in-place rescaling is the intended low-turnover mechanism and is correctly guarded (delta drift must exceed rebalance_band, never sells more than held, min_notional checked), but it makes the sandbox trade accounting noisy (avg_holding_bars 0.0, profit_factor 0.37, negative avg_trade_pnl_usd alongside a POSITIVE avg_trade_return_pct 0.64%) because partial-scale sells are counted as trade closes. This is a metrics artifact of continuous rescaling, not a code defect — but the optimizer/analyst should read per-trade stats with that in mind.

Developer's own honest caveat (echoed by the numbers): on full IWM history the vol-managed sleeve roughly MATCHES buy-and-hold Sharpe (~0.45 vs ~0.42) while cutting drawdown, i.e. the anomaly shows up as risk reduction, not a Sharpe lift. Sandbox Sharpe 0.11 (CI [-0.62, 0.86]), information_ratio -0.48, alpha -2.6% vs buy-hold, and long-only high beta (0.49) / benchmark_correlation 0.73 mean this is essentially a lightly-de-risked long IWM sleeve. It may struggle to clear the 0.5 OOS Sharpe floor even though it is correctly implemented. Not a correctness blocker; flagging for the analyst to judge the risk/return trade on full history.

Numerically viable full-history sample: 6,610 daily bars over 26 years, 37 trades (appropriate for a low-turnover long/flat overlay, not a broken signal), metrics_reliable=true.

Edge is present, not fee-dead: profit_factor 2.03 and avg_trade_return_pct 3.94% sit far above the USEQ ~0.05% floor and above L75's PF<1.2 'anomaly absent' tell — the vol-managed effect is real on IWM, it is just modest.

Trades match the hypothesis: long/flat continuous inverse-vol scaling, 0 shorts, exposure_pct 81%, recovery_factor 10.5, drawdown 36% (below the ~59% a raw buy-hold IWM took in 2008/2020) — the mechanism delivers the risk reduction it promises.

Low-parameter documented anomaly (Moreira-Muir) on a fresh instrument/venue (IWM.USEQ, 1.4% of corpus, zero commission) — structurally dodges the fee_edge and venue-concentration killers.

Beta-adjusted alpha field is positive (+1.2%), PSR 0.97, sharpe_ci_low marginally positive (0.03).

Closet-long risk: beta 0.49, benchmark_correlation 0.73, information_ratio -0.17 and QA's note that raw Sharpe (~0.41) barely matches buy-hold — the anomaly shows up as drawdown reduction, not a Sharpe lift.

Sharpe 0.41 is well below the 0.5 OOS validity floor; sharpe_ci_low is essentially 0, so this may fail deflation/holdout after optimization. This is the key thing the 3-phase run must resolve.

max_drawdown 36% and 5 losing calendar years — quality floors will be tight.

Commission-free USEQ venue and low-parameter documented academic mechanism structurally dodge the fee_edge and over-parameterization killers; sensitivity shows no cliffs.

Clean full-history sample (6,610 daily bars, 26 years) with metrics_reliable=true and per-trade return (1.88%) far above the USEQ fee floor.

Profit factor 1.65 and drawdown-reduction behaviour is real (the vol overlay does dampen exposure in hot-vol regimes).

Four HARD, non-waivable validity gates fail: OOS Sharpe 0.309 < 0.5, walk-forward is_overfitted=true (IS 0.77 -> OOS 0.309), PBO 0.60 > 0.5, holdout only 9 trades < 10 min. Promotion is impossible.

No alpha above beta: information_ratio -0.17, alpha 0.012, beta 0.49, benchmark_correlation 0.73 -- the result is a scaled-down long IWM position, not an active edge. It underperforms simply holding IWM on a risk-adjusted basis.

Not statistically significant: deflated_sharpe 0.454 (<0.95), is_significant=false, PSR consistent with no-skill after trial-count deflation, and does not survive programme-level FDR (4 of 278 kept, candidate does not survive).

Walk-forward OOS is inconsistent across windows (2.37 / 0.0 / -1.45) -- the positive average is carried by one window, not a stable edge.

The entire parameter sweep is uniformly mediocre (sensitivity Sharpe grid 0.27-0.34) with no robust high-edge region to tune toward; iterating parameters cannot raise an edge that is absent everywhere in the space.

Drawdown 33.4% breaches the hypothesis's own pre-registered 22% cap, and 3 of 5 pre-registered predictions missed (min_trades, min_oos_sharpe, max_drawdown).

Benjamini-Hochberg at q=0.10 over 278 programme candidates keeps 4. A candidate that does not survive here is not distinguishable from the programme's own noise, however good its individual statistics look.

VolatilityManagedIwmExposureOverlayU-e277145479

Applying the volatility-managed overlay to a single equity instrument reproduces scaled market beta rather than manufacturing alpha, and dampening drawdown does not create a statistically significant Sharpe premium when the entire parameter grid is uniformly mediocre.

The analyst abandoned it after optimization: four hard validity gates failed (OOS Sharpe < 0.5, is_overfitted=true, PBO > 0.5, holdout only 9 trades < 10) and it showed no alpha above beta, making it a scaled-down long IWM sleeve that underperformed simply holding IWM on a risk-adjusted basis.

Implement the Moreira-Muir volatility-managed portfolio anomaly on IWM.USEQ, continuously scaling a long/flat equity sleeve by the inverse of the ETF's own trailing realized volatility toward a self-calibrated vol target, on a commission-free venue with near-zero free parameters.

On 26 years of daily IWM data (6,610 bars) the initial backtest returned 299% with profit factor 2.03, Sharpe 0.41, 37 trades and a 36% max drawdown; after 3-phase optimization the OOS Sharpe was only 0.309 with walk-forward flagged overfitted (IS 0.77 -> OOS 0.309), PBO 0.60, deflated Sharpe 0.454 and information ratio -0.17.