The result
The hold-band strategy stays killed as a return-edge candidate. It reduced the largest historical decline and earned less than the comparison basket. The deeper study does not establish a superior risk-adjusted return. This is a negative result worth reporting plainly.
The rule and the comparison
The rule buys when an ETF closes above its 150-session moving average by 3%. It holds the position until the close falls the same distance below the average. The retained universe contains 12 ETFs: SPY, EFA, EEM, TLT, IEF, GLD, DBC, VNQ, XLE, XLK, XLU, XLV. These markets overlap. They are not independent bets.
The corrected model observes a closing signal, trades at the following close, and starts earning the next close-to-close return. It charges 7 basis points per equity transaction leg and includes daily rebalancing and initial entry costs. Uninvested capital earns a lagged Treasury yield proxy. That proxy is an estimate, not an investable cash return record.
The comparison basket holds the same ETFs at equal weights and rebalances daily. Calling it simple buy and hold would be inaccurate. Both portfolios use the same price snapshot and cost assumptions.
What the retained history shows
| Window | Annual return | Worst decline | Sharpe above cash |
|---|---|---|---|
| Full common history | 6.36% / 8.80% | -8.91% / -38.38% | 0.653 / 0.563 |
| 2013 onward | 6.02% / 9.01% | -8.91% / -26.26% | 0.649 / 0.660 |
| 2020 onward | 7.07% / 11.08% | -8.91% / -26.26% | 0.578 / 0.633 |
Annual return means compound annual growth. Worst decline means the largest peak-to-trough portfolio loss. Sharpe above cash measures average return after subtracting the same cash proxy, relative to volatility. The full common history runs from 2006-09-12 through 2026-10-27; the later slices start on 2013-01-02 and 2020-01-02. These are historical simulations, and the later slices are conditional research checks, not untouched validation.
Across the full history, the rule kept an average of 66.75% invested in ETFs. Lower exposure helps explain the smaller decline. It does not establish that timing added value.
Why the apparent advantage weakened
The first report used an idealized same-close fill assumption and a return-to-volatility ratio that did not subtract cash. The deeper study corrects execution timing and reports both cash conventions. From 2013 onward, the old-style ratio is 0.921 for the strategy and 0.817 for the basket. Subtracting cash produces the weaker strategy result shown in the table.
A passive stock-and-cash comparison sets its ETF weight from the earlier training period and freezes it. From 2013 onward, this mix earns 5.46% a year with a -13.62% worst decline and 0.659 Sharpe above cash. The active rule has no demonstrated timing advantage over this lower-exposure alternative.
The uncertainty still blocks the claim
The study resamples consecutive stretches of the paired historical return series to preserve some time dependence. With a mean block length of 63 sessions, its basic 95% confidence interval for the 2013 onward annual return difference is -6.191 to 0.286 percentage points. The corresponding Sharpe difference interval is -0.304 to 0.327. Both include zero.
The research record identifies at least 60 parameter choices. Earlier universe and method choices add selection risk that is not fully counted. The 2013 onward return difference is negative, and the statistical gate for a positive difference fails. The strategy recorded 140 completed sleeve round trips in that window, but overlapping ETF exposures prevent treating that count as independent evidence.
The sensitivity grid contains 48 settings. Only 8 beat the basket on Sharpe above cash. None beat it on compound annual return. The best new setting was not adopted. Choosing it after seeing the results would add another selection step.
What remains useful
The smaller historical decline is real within this simulation. It came with lower return. Neither the original SPY hold-band rule nor the separate drawdown filter passed the original acceptance gates. The broader follow-up also failed, and the corrected deep study gives no reason to revive a return-edge claim.
The retained prices are a retrospective adjusted-close snapshot. The ETF universe consists of survivors chosen after much of the history was visible. Opening fills, actual spreads, market impact, taxes and realized cash yields were not observed. Historical drawdown is not a future loss limit.
Keep the strategy killed. Keep the research record.
Any future paper test needs its universe, execution rule, cash treatment, costs, benchmark, trial ledger and success threshold fixed before unseen observations begin. No new test or service is activated by this report.