Executive Summary
This preliminary strategy-selection screen does not validate a live trading rule. It asks a narrower question: does XLP show enough evidence to justify chasing strength, buying weakness, or avoiding a standalone return signal?
Regime stability and a chain of autocorrelation/arima/garch tests converge on randomness; risk controls are the practical takeaway rather than a tradable edge.
Risk-managed, volatility-targeted, and regime-filtered rules should be treated as candidate tests until explicit out-of-sample backtests, transaction costs, turnover, and benchmarks are included.
Analysis Date And Sample Window
| Field | Value |
|---|---|
| Publication date | 2026-06-01 |
| Analysis run date | 2026-06-01 |
| Sample window | 2023-01-03 to 2024-12-30 |
| Return observations | 500 |
| Data fetched | 2026-05-10 13:14:33.644813 |
This page is the stable research page for this strategy screen. Normal reruns update this page so the main URL stays useful; separate dated editions are linked below only when we intentionally preserve a historical run.
How The Evidence Builds The Strategy View
The screen starts with the basic return profile, then tests whether daily returns tend to follow through or bounce. That order matters because a pleasant return profile is not the same thing as a tradable signal. The follow-through evidence is the main fork in the road for XLP: values below 1 in the variance-ratio test can lean toward mean reversion, but the p-values decide whether that lean deserves trust. The return-pattern model then checks whether a simple mean equation improves the story. When that model finds no useful structure, the practical bar rises: any proposed rule has to earn its keep in a direct backtest. Volatility evidence is still useful, but it answers a different question. It can support position sizing, risk limits, and stress checks; it does not turn weak return predictability into a standalone edge. The regime check keeps the conclusion honest. If the sample looks stable, regime filters belong in the next testing queue rather than in the conclusion.
Return Profile
| Metric | Value |
|---|---|
| Annualized return | 5.42% |
| Annualized volatility | 10.64% |
| Zero-rate Sharpe | 0.510 |
| Max drawdown | 13.57% |
| Lag-1 autocorrelation | -0.048 |
Zero-rate Sharpe means annualized return divided by annualized volatility. It is useful as a quick screen, but it is not a substitute for a benchmark-relative or risk-free-rate-adjusted evaluation.
Momentum Versus Mean Reversion
| Horizon | VR | HC_Statistic | Bootstrap_p | Reject_Random_Walk |
|---|---|---|---|---|
| VR q=2 | 0.951 | -1.111 | 0.300 | No |
| VR q=4 | 0.855 | -1.783 | 0.092 | No |
| VR q=8 | 0.863 | -1.067 | 0.314 | No |
| VR q=16 | 0.930 | -0.446 | 0.856 | No |
Values below 1 can lean toward bounce behavior, while values above 1 can lean toward follow-through. Here, the variance-ratio values are not strong enough to reject the random-walk null, so the return signal remains too weak to trust as a standalone rule.
Stationarity Check
| Test | P_Value |
|---|---|
| ADF returns | 0.0100 |
| KPSS returns | 0.1000 |
| Phillips-Perron returns | 0.0100 |
These tests check whether the return series is usable for time-series modeling. They do not create a signal by themselves.
Return-Pattern Model
| Metric | Value |
|---|---|
| ARIMA order | (0,0,0) |
| ARFIMA d median | 0.068 |
| Residual Ljung-Box p | 0.6622 |
| Squared-residual Ljung-Box p | 0.0058 |
| Model conclusion | short_memory |
The mean-equation model is a confirmation step. If it does not find a useful return structure, the burden shifts to explicit strategy backtests rather than narrative conviction.
Volatility Clustering
ARCH-effects conclusion: weak_arch_effects.
Volatility evidence can support risk controls, but it does not create a return-prediction edge.
Volatility Model
| Metric | Value |
|---|---|
| Best volatility model | sGARCH (norm) |
| Persistence | 0.999 |
| Half-life | 558.690 trading days |
| Squared standardized residual p | 0.0151 |
Volatility modeling is useful for candidate risk controls. If residual diagnostics remain imperfect, treat the result as a prompt for further testing, not a finished risk model.
Regime Context
Regime conclusion: stable. Detected structural breakpoints: 0.
If the sample is stable, regime filters should be framed as candidate tests rather than evidence-backed conclusions.
Candidate Strategy Hypothesis
The evidence supports a research hypothesis, not a live rule: test whether a risk-aware allocation process adds value when the return signal itself is weak.
{
"strategy_name": "XLP Risk-Aware Allocation Test",
"strategy_status": "hypothesis_for_backtest",
"strategy_type": "risk_managed_allocation",
"asset": "XLP",
"core_thesis": "Return predictability is weak, so any practical rule should be tested through explicit risk controls rather than assumed momentum or mean reversion.",
"required_backtests": ["walk-forward validation", "buy-and-hold asset benchmark", "broad market benchmark", "cash or T-bill benchmark", "transaction costs", "turnover"],
"not_investment_advice": true
}Backtested Results
The downloadable backtested results are planned for a later implementation step. They should include walk-forward results, benchmarks, turnover, and transaction-cost sensitivity before any rule is treated as validated.
Research Run History
| Run | Sample | Status | Link |
|---|---|---|---|
| 2026-06-01 | 2023-01-03 to 2024-12-30 | Current stable-page analysis | This page |
Separate dated research runs will be linked here when we publish a historical edition rather than updating the stable page.
Limitations
This is a preliminary strategy-selection screen based on precomputed research outputs. It is not personalized financial advice and it is not a production trading rule.