Capture Timing v4 Multi-Period
19 Rolling PeriodsTesting the same 12 dividend capture strategies across 19 rolling 3-month periods to measure consistency and robustness across different market conditions (Jan 2024 - Sep 2025).
What's Wrong With This Experiment
This experiment has critical data quality issues including potential look-ahead bias from incorrect declaration dates. Results should be treated as indicative only and not used for trading decisions.
Look-Ahead Bias: Incorrect Declaration Dates
A significant percentage of positions may have entered before the dividend was publicly declared. The declaration_date data from the API is unreliable - some dates are recorded months after the actual announcement.
Impact: Returns may be inflated by trades that used future knowledge of dividend announcements.
Missing Declaration Dates Not Excluded
Positions with NULL declaration dates were not excluded from the experiment. Per best practices, these should have been excluded since we cannot verify they were publicly announced before entry.
Impact: These positions may include additional look-ahead bias that cannot be quantified.
Overlapping 3-Month Periods
The 19 rolling 3-month periods overlap by 2 months each. This means consecutive periods share ~67% of their entry window, making them correlated and not independent samples.
Impact: Period-by-period consistency statistics may be misleading.
No Transaction Costs
Results do not include commissions, bid-ask spreads, or market impact. High-turnover strategies would be most affected.
Impact: Real-world returns would be significantly lower than reported.
Potential Survivorship Bias
The stock universe may be biased toward currently active stocks, excluding companies that were delisted during the experiment period.
Impact: Universe may be skewed toward successful companies.
Quality Score Filter Applied
Stocks were filtered by a quality_score >= 50 from the legacy scoring system. This pre-selection may have biased the universe.
Impact: Results may not generalize to the broader dividend stock universe.
Note: These issues have been identified through post-hoc audit. Future experiments will address these problems through improved data validation, stricter look-ahead bias checks, and better methodology. This experiment (V4 Multi-Period) will not be re-run but serves as a learning reference.
Comparison: v3 Single-Period vs v4 Multi-Period Average
v3 tested a single 3-month period (Oct 2025 - Jan 2026). v4 tests 19 overlapping periods.
| Strategy | v3 Return | v4 Avg (19 periods) | v4 Std Dev | Difference |
|---|---|---|---|---|
| 1D → BE | +10.36% | +5.34% | 5.46% | -5.02pp |
| 7D → BE | +9.54% | +4.78% | 4.32% | -4.76pp |
| 7D → 1D | +10.58% | +4.51% | 4.68% | -6.07pp |
| 1D → 14D | -1.52% | +4.15% | 3.85% | +5.67pp |
| 7D → 7D | +19.46% | +3.65% | 3.95% | -15.81pp |
| 7D → 14D | +2.42% | +3.61% | 3.46% | +1.19pp |
| 14D → 7D | +2.81% | +3.23% | 3.05% | +0.43pp |
| 1D → 1D | +8.77% | +3.13% | 3.68% | -5.64pp |
| 1D → 7D | +2.09% | +2.34% | 4.13% | +0.25pp |
| 14D → 1D | +2.19% | +2.23% | 3.19% | +0.04pp |
Period Performance Heatmap
Total return by strategy (rows) and period (columns). Darker green = higher returns.
Strategy Performance Summary
Aggregated statistics across all 19 periods
| Strategy | Avg Return | Std Dev | Min | Max | % Profitable | Win Rate | Sharpe |
|---|---|---|---|---|---|---|---|
1D → BE 111 avg trades/period | +5.34% | 5.46% | -6.78% | +15.26% | 79% | 82.8% | 0.98 |
7D → BE 77 avg trades/period | +4.78% | 4.32% | -3.78% | +12.62% | 95% | 78.5% | 1.11 |
7D → 1D 131 avg trades/period | +4.51% | 4.68% | -4.42% | +11.27% | 84% | 42.5% | 0.96 |
1D → 14D 83 avg trades/period | +4.15% | 3.85% | -3.10% | +13.34% | 89% | 59.6% | 1.08 |
7D → 7D 85 avg trades/period | +3.65% | 3.95% | -1.32% | +12.36% | 74% | 55.6% | 0.92 |
7D → 14D 62 avg trades/period | +3.61% | 3.46% | -1.88% | +12.11% | 95% | 58.8% | 1.04 |
14D → 7D 62 avg trades/period | +3.23% | 3.05% | -5.47% | +8.23% | 95% | 64.0% | 1.06 |
14D → BE 56 avg trades/period | +3.14% | 3.11% | -3.99% | +7.88% | 84% | 76.9% | 1.01 |
1D → 1D 283 avg trades/period | +3.13% | 3.68% | -6.15% | +10.30% | 84% | 33.0% | 0.85 |
1D → 7D 141 avg trades/period | +2.34% | 4.13% | -6.39% | +9.59% | 74% | 51.9% | 0.57 |
14D → 1D 70 avg trades/period | +2.23% | 3.19% | -6.84% | +6.18% | 84% | 49.8% | 0.70 |
14D → 14D 47 avg trades/period | +1.37% | 2.64% | -4.64% | +6.31% | 79% | 65.4% | 0.52 |
Key Insights
1D → 14D leads in returns: With an average return of +5.34%, the 1D entry / 14D exit strategy produced the highest returns. However, it also has high volatility (std dev: 5.5%) indicating inconsistent performance.
1D → 1D is most consistent: Despite lower returns, the 14D → 14D strategy has the lowest standard deviation (2.6%) and was profitable in79% of periods, making it more predictable.
Breakeven strategies show moderate performance: The BE (breakeven) exit strategies cluster in the middle of the pack, suggesting that waiting for price recovery after ex-date provides steady but not exceptional returns.
Early 2024 was exceptional: Period 1 (Jan-Mar 2024) showed unusually high returns across multiple strategies, possibly due to favorable market conditions. This highlights the importance of multi-period testing.
How Stocks Were Picked (Medium Opportunities)
This experiment targets "Medium Opportunity" stocks, defined as those with a Quality Score ≥ 50. This filters out yield traps and highly speculative plays while retaining a broad universe of investable assets.
Selection Criteria
- Dividend Consistency: Generally 3+ years of reliable payments.
- Growth: Positive 5-year Dividend CAGR or growth streaks.
- Safety: Payout ratios typically under 60% and covered by Free Cash Flow.
- Value: Reasonable valuation metrics (P/E, Yield Percentile).
Exclusions
- Yield Traps: Stocks with >50% annual yield.
- Extreme Risk: Stocks with single payments >20% of share price.
- Data Gaps: Missing price history or fundamental data.
Caveats & Methodology Notes
Rolling periods overlap: Each 3-month period starts 1 month after the previous one, so periods share 2 months of data. This provides smoother estimates but means results are not fully independent.
Extended exits: Unlike v3, positions in v4 can exit after the entry window ends (up to 14 days for fixed exit, 90 days for breakeven). This allows natural trade completion.
No transaction costs: Results do not include commissions, spreads, or market impact. High-turnover strategies like 1D → 1D would be most affected by real-world costs.
Related Experiments
Compare with single-period experiments for additional context.
Disclaimer: This is a paper-trading experiment for research purposes only. Past performance does not guarantee future results. Transaction costs were not simulated. Results may not be reproducible in live trading due to slippage, liquidity constraints, and market impact.