Continuous Macro-Stress Smooth-Score Growth-vs-Defensive ETF Style Allocation

Strategy Card

Continuous Macro-Stress Smooth-Score Growth-vs-Defensive ETF Style Allocation — strategy card

Spec ID: spec-continuous-macro-stress-style-timing-us-etfs-1786122720 · Generated: 2026-08-07 17:54 UTC

Cluster: Tactical Allocation · Sub Cluster: Macro-Stress Growth Defensive Etf Timing

One-line description

Dynamically tilts a U.S. ETF portfolio between growth/technology and defensive/value-income baskets using a continuous macro-stress score built from rates, equity drawdown, VIX regime, VIX relief, and recent growth extension. The score is transformed through a tanh tilt and EWMA-smoothed before rebalancing.

Why this trade exists

The trade expresses the idea that growth-vs-defensive leadership is partly state-dependent. Falling long rates, equity drawdown relief, and volatility stress relief can improve the relative payoff to long-duration growth assets, while crowded growth performance in low-volatility/rate-quiet regimes can make defensive/value-income exposure more attractive.

The strategy does not forecast single-name alpha. It rotates between diversified ETF style baskets, aiming to capture broad regime changes while avoiding binary all-in/all-out timing through continuous scoring and slow weight smoothing.

Algorithm

[code omitted from public view]

Parameters

Param	Value	Notes
Strategy variant	Main selected smooth-score policy	No optional bond/credit overlay.
Growth basket	QQQ, XLK, VGT, SPYG, VUG	Equal-weight within basket.
Defensive basket	SCHD, VYM, VTV, FDVV, COWZ	Equal-weight within basket.
Signal-only ETF	SPY	Used for drawdown feature; not part of traded basket in the target allocation.
Macro inputs	DGS10, VIX	10Y yield and volatility regime/relief inputs.
Backtest window	2020-01-01 to 2025-10-08	Daily bars.
alpha	0.50	Blend between rate relief and SPY drawdown core score.
lambda_s	0.50	Weight on stress interaction score.
lambda_c	0.05	Penalty on crowded growth/low-volatility score.
MaxTilt	0.50	Growth weight range is theoretically 0% to 100% before any platform position caps.
tau_w	0.75	Tanh score-temperature for target weight.
eta	0.05	EWMA smoothing speed for actual growth weight.
softplus_tau	1.00	Softplus smoothing parameter.
Rebalance	Daily close	Continuous target update and close rebalance convention.
Trading cost assumption	10 bps paper turnover-cost parameter; platform costs include $0.004/share commission, $1 minimum, 1% max commission, 0 bps slippage	The code notes cost_bps is parsed for paper fidelity and may not be applied as explicit trade slippage/commission in generated orders.
Price policy	Real ETF closes for execution	Short missing signal gaps may be forward-filled for analysis features only, not for execution prices.

Look-ahead audit

#	Concern	Status
1	Close-derived ETF, SPY, VIX, and DGS10 features	Uses information through the signal date only; no future returns are required for feature construction.
2	Execution timing	Rebalances at the close under the platform close/MOC convention; no same-day open execution from same-day close signals is specified.
3	Expanding z-scores and running peaks	Expanding calculations are causal and initialized with available warmup history only.
4	Macro data availability	DGS10 uses an observation-date point-in-time filter; release-time lag risk remains if the database timestamp is only date-level.
5	Missing prices	Forward-fill is limited to analysis signal construction; execution prices must be real closes or trades are skipped.

Caveats / known limitations

• The backtest window is 2020-01-01 to 2025-10-08, not the paper's 2017-06-28 to 2026-05-15 main aligned window; all paper performance predictions tied to the original main, OOS, and post-2022 windows are reference diagnostics rather than descriptions of this run.
• The paper-fidelity notes include an undescribed declared deviation that invalidates the paper's factor-attribution predictions for G-D beta exposures, annualized alpha, Newey-West t-statistic, and adjusted R²; this card should not treat those factor numbers as verified for the traded implementation.
• The fixed-parameter validation predictions over 2018-06-28 to 2026-05-15 do not apply to this selected main-policy run over the task window.
• The optional strict incremental bond/credit overlay is not implemented, so the paper's Old Best + Bond/Credit main-window, OOS, post-2022, and cost-sensitivity predicted results do not describe this strategy.
• If the platform enforces the 10% per-ETF cap as a hard cap, extreme 50% tilt states cannot exactly replicate the intended equal-basket paper weights; the resulting weight path and performance can differ from the source design.
• Our run only partly resembles the paper's selected-policy diagnostics: max drawdown is very close to the cited -31.63%, but Sharpe is lower at 0.92 and volatility is higher at 21.08%, plausibly due to the different 2020-2025 sample, implementation caps, and platform cost model.
• The code indicates the 10 bps cost_bps parameter is parsed for paper fidelity while returned trades do not directly include that turnover-cost formula; realized costs may therefore differ from the paper's cost convention.
• The strategy is a style-timing allocator with high trade count, not a standalone alpha model; performance remains exposed to U.S. equity beta, technology/growth concentration, rate-regime instability, and ETF basket construction choices.

Results

The backtest produced a positive total return of 111.24% with Sharpe 0.92, Sortino 1.20, and beta to SPY of 0.76. Drawdown control was broadly in line with the paper-style objective, with max drawdown of -31.61%, but realized volatility was still equity-like at 21.08%. The high win rate and profit factor are favorable diagnostics, while the 13,932 trades highlight meaningful implementation and cost sensitivity.

Run diagnostics

Recomputed here from the run's stored daily returns, because every annualized figure in the table above is derived from the LENGTH of that array (years = n / 252). Only total return and max drawdown do not depend on the row count, so when the recomputed and stored values differ by a common factor, those two are the numbers to trust.

Check	From the run's own returns	Note
Daily-return rows	1,450	1.00x the 1,454 trading days in 2020-01-01..2025-10-08
Distinct dates	1,450	one row per date
Date span	2020-01-02 .. 2025-10-08
Sum of daily returns	111.24%	matches the reported total return
Sharpe from these rows	0.92	stored 0.92
Volatility from these rows	21.08%	stored 21.08%
Max drawdown from these rows	-31.61%	stored -31.61%
CAGR from these rows	13.88%	stored 13.88%

Backtest metrics snapshot

Metric	Value
Total Return	111.24%
Sharpe	0.92
Sortino	1.20
Calmar	0.44
Max Drawdown	-31.61%
Volatility	21.08%
Beta vs SPY	0.76
Win Rate	83.00%
Profit Factor	11.83
Total Trades	13932
Symbols	11 (COWZ, FDVV, QQQ, SCHD, SPY, SPYG, VGT, VTV, VUG, VYM, +1 more)