I want to test a simple question. Can an annually rebalanced portfolio of companies that grow their dividends per share and reduce shares outstanding beat SPY after trading costs? Dividends put cash in shareholders’ hands. When the share count falls, investors who keep the same number of shares own a larger fraction of the company. This strategy brings both outcomes into one stock ranking.
Universe and rebalancing
At the close of the final trading day in May, screen the historical universe of US-listed, USD-denominated common stocks for company market capitalizations strictly above US$10 billion. Keep one representative share class per company. The default holds the top 30 names, equal-weighted at a 98% target equity allocation, with a 2% cash buffer. Rebalance at the next session’s open. If fewer than 30 names qualify for scoring, equal-weight the eligible names.
The prototype assumes 10 basis points each for fees and slippage per side. The SPY benchmark buys on the same day with the same initial equity allocation and trading costs, then holds. Both use total-return series with dividends reinvested.
The shareholder return score
| Factor | Definition | Default | October reference |
|---|---|---|---|
| Dividend yield | Latest available full-year regular dividends per share (DPS) divided by the signal-day close on a consistent share basis | 25% | 0% |
| DPS growth | Eight-year compound annual growth in regular DPS | 30% | 40% |
| Net share reduction | Annualized eight-year reduction in split-adjusted fiscal year-end shares outstanding | 30% | 40% |
| Reduction acceleration | Average of the latest three annual net share-reduction rates minus the earliest three | 15% | 20% |
Convert each factor to a percentile among eligible stocks that year and sum the weighted percentiles. For each regular DPS cut strictly greater than 20% in the latest three annual comparisons, multiply the score by 0.85. Two cuts give a multiplier of 0.85². The October reference also applies a 0.95 multiplier for each year-over-year drop in actual buyback cash strictly greater than 50% over the same window.
Eight years of growth require nine consecutive fiscal-year observations. Regular DPS excludes special dividends. Missing values stay blank, and a confirmed zero starting DPS receives zero points for the growth factor. Net share reduction reflects buybacks, issuance, stock compensation and share-funded acquisitions. Both weight sets are research hypotheses.
What was knowable at the time
Each rebalance uses financial records that had been disclosed and were available from the data source by the signal cutoff. The universe is rebuilt from each year’s historical snapshot, including companies that later delisted, failed or were acquired. Their exit settlements also matter to the portfolio’s realized return.
What the prototype has verified
The vectorbt prototype and Bloomberg Excel/CSV input template are complete. The 4 November 2026 validation log records 31 passing tests on synthetic data, including late financial revisions, the dividend-cut threshold and cash settlements for delistings. Historical returns and drawdowns remain to be measured. Zero-recovery bankruptcies, valuations during trading halts and stock-for-stock mergers also need additional event handling.
Once the full history is available, the next step is to reconcile the annual universe, the financial-data timestamps and the total-return opening prices. Then compare both specifications on returns relative to SPY after costs, maximum drawdown and turnover.
vectorbt order documentation · Bloomberg point-in-time fundamentals