TL;DR
Quality filters helped, but the sixty-configuration search added little over the unchanged value portfolio. Across four restarted tests from 2018 through September 2026, selection returned 288.53% versus 285.39% for the unchanged portfolio.
The unchanged portfolio also won the January 2025 through September 2026 comparison: 60.31% versus the training winner’s 36.19%. The shortlist overlapped early test years, so these are retrospective comparisons rather than evidence from unseen dates.
The broad search added only 3.14 points over the unchanged value portfolio
Portfolio value (USD)
- Retrospective selection procedure 2026-09-30, 16:00:00 EDT
- $38,853.05
- Unchanged cash/operating-profit portfolio 2026-09-30, 16:00:00 EDT
- $38,538.58
- Cheap stocks only 2026-09-30, 16:00:00 EDT
- $18,449.37
- SPY: reinvested dividends, initially 95% stocks 2026-09-30, 16:00:00 EDT
- $30,344.94
Retrospective selection procedure: 27.58% largest decline in plotted observations. 2021-12-27, 16:00:00 EST ($20,955.69) to 2022-09-26, 16:00:00 EDT ($15,176.47).
Show the table · 6 rows, 4 columns
Scroll sideways to compare all columns. Select a heading to sort.
| Retrospective selection procedure | +288.53% | 28.04% | $38,853.05 |
| Unchanged cash/operating-profit portfolio | +285.39% | 29.63% | $38,538.58 |
| Cheap stocks only | +84.49% | 63.71% | $18,449.37 |
| SPY: cash dividends, initially 95% stocks | +197.71% | 32.49% | $29,771.17 |
| SPY: full exposure, reinvested dividends, fee-free context | +219.78% | 34.10% | $31,978.06 |
| SPY: reinvested dividends, initially 95% stocks | +203.45% | 32.49% | $30,344.94 |
No matching rows. Clear the filter to see all records.
Fold returns compound rather than add. Every fold starts from cash; all controls use the same boundary convention. Full-exposure SPY is fee-free context.
Chaining four retrospective tests that restarted each portfolio from cash, the selection procedure returned +288.53%. The unchanged value portfolio returned +285.39%; the reinvested SPY control starting each test at 95% stocks returned +203.45%. Fully invested, fee-free SPY context returned +219.78%. The tests cover 2018 to September 2026.
The earlier equal-weight value portfolio still returned +60.31% in January 2025 to September 2026. The new training leader returned +36.19% over the same dates. A bigger search found different useful rules, but it added only 3.14 percentage points over the unchanged incumbent in the retrospective reset-fold chain.
The candidate shortlist used dates that overlap early fold tests. This is retrospective research, not an untouched future test.
Retrospective reset-fold drawdowns: 2018 to September 2026
Value (%)
- Retrospective selection procedure 2026-09-30, 16:00:00 EDT
- -10.2%
- Unchanged cash/operating-profit portfolio 2026-09-30, 16:00:00 EDT
- -8.02%
- Cheap stocks only 2026-09-30, 16:00:00 EDT
- -10.51%
- SPY: reinvested dividends, initially 95% stocks 2026-09-30, 16:00:00 EDT
- -1.9%
Inspect reset-fold loss paths
Every test begins with cash and $10,000, then its end NAV/initial capital compounds into the next fold. This includes entry fees. The displayed chain does not carry positions continuously between folds.
Start with cheapness plus business-quality filters
The unchanged value portfolio buys ten low-P/E stocks with positive cash flow and operating income, and net income/assets above 5%. Equal weights and a 90-day refresh keep the comparison simple. It remained stronger than the new training winner on the most recent dates.
Twenty selection ideas, sixty starting portfolios
Inspect all twenty selection families
We changed what cheapness means, how business quality is measured, and which screen binds first. Earnings multiples, sales multiples, book value, free cash flow and enterprise value answer different questions. Every family also required a positive P/E so earnings-yield weights remained meaningful.
Equal weights spread money evenly. Earnings-yield weights give more money to higher 1/P/E scores. Native risk parity uses the selected stocks' historical covariance; it can retain equal-weight fallback for stocks with insufficient history. All three methods target the same 95% stock deployment.
Scroll sideways to compare all columns. Select a heading to sort.
| Income/assets with P/E below 25 | +516.58% / 26.39% | +478.19% / 29.95% | +524.84% / 23.87% |
| Retained cheap-value income/assets 5% cash and operating filters | +416.41% / 27.23% | +468.99% / 28.53% | +402.51% / 26.25% |
| Highest operating income/enterprise value | +227.33% / 39.66% | +389.72% / 41.40% | +200.90% / 41.04% |
| Free cash flow margin with sales multiple below 3 | +366.26% / 43.42% | +275.48% / 47.31% | +347.27% / 39.36% |
| Cheap book with positive equity and income | +364.76% / 50.19% | +267.85% / 51.86% | +335.59% / 47.29% |
| Highest 40 cash-flow yields, then price momentum | +341.33% / 43.28% | +252.22% / 47.05% | +324.03% / 42.21% |
| Highest EBITDA/enterprise value | +208.54% / 44.10% | +329.11% / 45.79% | +200.33% / 44.85% |
| Operating margin with positive cash flow | +302.10% / 31.62% | +320.45% / 31.98% | +321.74% / 34.59% |
| Income/equity with book multiple below 4 | +287.65% / 47.78% | +279.63% / 48.38% | +321.50% / 43.48% |
| Operating income/assets with sales multiple below 3 | +274.47% / 33.33% | +317.12% / 33.93% | +298.64% / 31.97% |
| Strongest 40 income/assets, then cheapest | +279.56% / 30.04% | +299.88% / 30.84% | +275.04% / 30.08% |
| Highest free cash flow yield | +290.40% / 55.21% | +298.32% / 56.55% | +276.71% / 53.56% |
| Gross profit/assets with operating profit | +288.85% / 33.82% | +277.71% / 34.97% | +294.88% / 33.65% |
| Lowest income-minus-cash-flow/assets | +246.10% / 34.92% | +247.45% / 36.92% | +254.26% / 35.81% |
| Cheap sales with operating profit | +221.19% / 49.23% | +218.21% / 54.11% | +180.63% / 46.20% |
| Income growth with P/E below 25 | +157.53% / 45.20% | +221.07% / 47.21% | +153.39% / 41.86% |
| Cheapest 40, then strongest income/assets | +190.96% / 45.86% | +205.67% / 46.44% | +177.79% / 48.38% |
| Revenue growth with P/E below 25 | +132.02% / 42.92% | +104.73% / 45.47% | +173.85% / 34.59% |
| Lowest positive P/E | +80.43% / 60.67% | +107.13% / 58.02% | +83.52% / 59.66% |
| Earnings yield with debt/assets below 30% | +85.34% / 59.33% | +98.82% / 57.10% | +100.45% / 59.42% |
No matching rows. Clear the filter to see all records.
All sixty native configurations completed on January 2011 to December 2021. Each cell shows total return after modeled fees, then native maximum drawdown. These dates selected the rules.
A strong training score did not stay strongest
Portfolio value (USD)
- Training leader: income/assets with risk parity 2021-12-31, 16:00:00 EST
- $62,483.66
- Unchanged cash/operating-profit portfolio 2021-12-31, 16:00:00 EST
- $51,640.54
- Cheap stocks only 2021-12-31, 16:00:00 EST
- $18,042.90
- SPY: reinvested dividends, initially 95% stocks 2021-12-31, 16:00:00 EST
- $44,175.35
Training leader: income/assets with risk parity: 23.76% largest decline in plotted observations. 2020-02-20, 21:10:56 EST ($37,066.49) to 2020-03-23, 18:58:34 EDT ($28,259.64).
One highest-return allocator per economic family, with the cheap-only baseline and reinvested SPY control included. The complete rule and benchmark names are shown in the adjoining tables.
Each line starts at $10,000. The chart keeps the last native observation for each New York date; tables report maximum drawdown from the complete native path. Initial purchase fees remain in the return.
Training drawdowns: 2011 to 2021
Value (%)
- Training leader: income/assets with risk parity 2021-12-31, 16:00:00 EST
- -3.04%
- Unchanged cash/operating-profit portfolio 2021-12-31, 16:00:00 EST
- -2.62%
- Cheap stocks only 2021-12-31, 16:00:00 EST
- -3.54%
- SPY: reinvested dividends, initially 95% stocks 2021-12-31, 16:00:00 EST
- -0.52%
Inspect the training drawdowns
Negative values show the loss from the running high of the displayed daily NAV observations. Sampling can make the chart extreme slightly smaller than the native maximum in the tables.
The order of the screens mattered
Taking the strongest forty income/assets scores before buying the ten cheapest stocks returned +279.56% with 30.04% drawdown. Taking the forty cheapest first, then buying the ten strongest income/assets scores returned +190.96% with 45.86% drawdown. Both used equal weights and the same eligible universe.
The first screen decides which companies can reach the final ranking. This comparison favored starting with quality. Pure free-cash-flow yield returned +290.40% with 55.21% drawdown, while revenue growth under a reasonable P/E returned +132.02% with 42.92% drawdown. A cash or growth label alone did not match the strongest profitability family.
More concentration and faster refresh did not improve the leaders
Eight additional configurations tested five or twenty stocks with a 45-day or 180-day refresh, using the best allocator from each of the two strongest distinct families. None exceeded the original income/assets leader at +524.84%. Its five-stock, 45-day version fell to +381.05%, with 36.42% drawdown and $1,047.78 in modeled stock fees.
The original ten-stock, 90-day rule remained strongest. The twenty-stock, 180-day version reduced drawdown to 22.99%, but its return fell to +446.80%. That lower drawdown is a real tradeoff rather than a return improvement.
Show the table · 8 rows, 6 columns
Scroll sideways to compare all columns. Select a heading to sort.
| Income/assets, risk parity | 5 | 45 days | +381.05% | 36.42% | $1,047.78 |
| Income/assets, risk parity | 5 | 180 days | +477.37% | 25.01% | $461.75 |
| Income/assets, risk parity | 20 | 45 days | +370.56% | 29.79% | $727.71 |
| Income/assets, risk parity | 20 | 180 days | +446.80% | 22.99% | $315.73 |
| Cash/operating filters, earnings yield | 5 | 45 days | +463.28% | 33.37% | $956.47 |
| Cash/operating filters, earnings yield | 5 | 180 days | +277.61% | 34.84% | $299.96 |
| Cash/operating filters, earnings yield | 20 | 45 days | +427.02% | 29.14% | $740.58 |
| Cash/operating filters, earnings yield | 20 | 180 days | +329.20% | 28.24% | $290.15 |
No matching rows. Clear the filter to see all records.
January 2011 to December 2021; same $10,000 starting capital and 0.05% fee per stock fill.
Cash and valuation safeguards produced mixed results
Adding positive free cash flow to the training leader lowered its return to +514.50%. Adding liabilities/assets below 75% lowered it to +323.41%. The incumbent with earnings-yield weights improved from +468.99% to +480.40% after adding P/E below 35.
One declared cash cell added income/assets above 3% to a portfolio already requiring more than 5%. Its +468.99% result reproduced the original because that weaker requirement did not change eligibility. It is retained as a redundant cell, not a separate economic improvement.
Show the table · 4 rows, 3 columns
Scroll sideways to compare all columns. Select a heading to sort.
| Income/assets leader: add positive free cash flow | +514.50% | 26.14% |
| Income/assets leader: add liabilities/assets below 75% | +323.41% | 27.08% |
| Incumbent earnings yield: redundant 3% income/assets check | +468.99% | 28.53% |
| Incumbent earnings yield: add P/E below 35 | +480.40% | 28.53% |
No matching rows. Clear the filter to see all records.
Four declared safeguard cells after the eight breadth/refresh cells. All 72 requested training cells completed; one safeguard is redundant.
The same eight candidates across four retrospective folds
Show the table · 4 rows, 8 columns
Scroll sideways to compare all columns. Select a heading to sort.
| 2011 to 2017 | 2018 to 2019-12-31 | Income/assets + risk parity | +201.02% | +14.69% | 18.33% | +15.30% | +22.86% |
| 2011 to 2019 | 2020 to 2021-12-31 | Income/assets + risk parity | +246.26% | +80.77% | 23.82% | +59.93% | +48.18% |
| 2011 to 2021 | 2022 to 2023-12-31 | Income/assets + risk parity | +524.84% | +3.13% | 27.27% | +7.55% | +2.31% |
| 2011 to 2023 | 2024 to 2026-09-30 | Incumbent + earnings yield + P/E below 35 | +614.81% | +81.72% | 15.44% | +94.32% | +62.93% |
No matching rows. Clear the filter to see all records.
All dates were known before this campaign. The shortlist search overlaps early test years; this is a retrospective comparison, not independent untouched evidence.
The shortlist contained the six highest-return distinct economic families, the unchanged equal-weight incumbent and cheap-only equal weights. It stayed fixed across all four folds. Each fold chose the highest earlier-period total return, then evaluated those unchanged rules in the next period.
Income/assets with risk parity won the first three training selections. After training through 2023, the incumbent with earnings-yield weights and a P/E ceiling won the fourth. The selected rules lagged SPY in 2018 to 2019. In 2024 to September 2026, they beat SPY but lagged both the unchanged incumbent and cheap-only.
The training leader lagged both controls on 2022 to 2024 dates
Portfolio value (USD)
- Training leader: income/assets with risk parity 2024-12-31, 16:00:00 EST
- $11,590.33
- Unchanged cash/operating-profit portfolio 2024-12-31, 16:00:00 EST
- $12,780.99
- Cheap stocks only 2024-12-31, 16:00:00 EST
- $12,225.10
- SPY: reinvested dividends, initially 95% stocks 2024-12-31, 16:00:00 EST
- $12,661.18
Training leader: income/assets with risk parity: 26.80% largest decline in plotted observations. 2022-01-02, 19:00:00 EST ($10,000.00) to 2022-09-26, 16:00:00 EDT ($7,320.11).
Show the table · 6 rows, 3 columns
Scroll sideways to compare all columns. Select a heading to sort.
| Cheap stocks only | +22.25% | 26.67% |
| Training leader: income/assets with risk parity | +15.90% | 27.27% |
| Unchanged cash/operating-profit portfolio | +27.81% | 29.29% |
| SPY: cash dividends, initially 95% stocks | +25.45% | 25.21% |
| SPY: reinvested dividends, initially 95% stocks | +26.61% | 25.30% |
| SPY: full exposure, reinvested dividends, fee-free context | +28.43% | 25.06% (comparison curve) |
No matching rows. Clear the filter to see all records.
Same dates and capital. Stock-book and 95% SPY statistics use modeled stock fees; full-exposure SPY is fee-free context with drawdown measured from its complete native comparison curve. Basket cash dividends can be redeployed at refresh.
Each line starts at $10,000. The chart keeps the last native observation for each New York date; tables report maximum drawdown from the complete native path. Initial purchase fees remain in the return.
The unchanged training leader on 2022 to 2024: drawdowns
Value (%)
- Training leader: income/assets with risk parity 2024-12-31, 16:00:00 EST
- -3.92%
- Unchanged cash/operating-profit portfolio 2024-12-31, 16:00:00 EST
- -4.07%
- Cheap stocks only 2024-12-31, 16:00:00 EST
- -6.88%
- SPY: reinvested dividends, initially 95% stocks 2024-12-31, 16:00:00 EST
- -3.44%
Inspect 2022 to 2024 loss paths
Negative values show the loss from the running high of the displayed daily NAV observations. Sampling can make the chart extreme slightly smaller than the native maximum in the tables.
The original value screen still won the most recent comparison
Portfolio value (USD)
- Training leader: income/assets with risk parity 2026-09-30, 16:00:00 EDT
- $13,619.32
- Unchanged cash/operating-profit portfolio 2026-09-30, 16:00:00 EDT
- $16,031.23
- Cheap stocks only 2026-09-30, 16:00:00 EDT
- $15,268.27
- SPY: reinvested dividends, initially 95% stocks 2026-09-30, 16:00:00 EDT
- $12,991.19
Training leader: income/assets with risk parity: 16.84% largest decline in plotted observations. 2026-01-08, 16:00:00 EST ($13,107.57) to 2026-03-27, 17:13:30 EDT ($10,900.83).
Show the table · 6 rows, 3 columns
Scroll sideways to compare all columns. Select a heading to sort.
| Cheap stocks only | +52.68% | 22.82% |
| Training leader: income/assets with risk parity | +36.19% | 16.99% |
| Unchanged cash/operating-profit portfolio | +60.31% | 15.64% |
| SPY: cash dividends, initially 95% stocks | +29.65% | 19.22% |
| SPY: reinvested dividends, initially 95% stocks | +29.91% | 19.22% |
| SPY: full exposure, reinvested dividends, fee-free context | +31.54% | 20.19% (comparison curve) |
No matching rows. Clear the filter to see all records.
Same dates and capital. Stock-book and 95% SPY statistics use modeled stock fees; full-exposure SPY is fee-free context with drawdown measured from its complete native comparison curve. Basket cash dividends can be redeployed at refresh.
The earlier research used 2016 to 2021 to choose the lowest positive P/E portfolio with available net income/assets above 5%, positive free cash flow and positive operating income. It selected up to ten equal weights with a 90-day refresh. Its final 2025 to September 2026 comparison returned +60.31%; cheap-only returned +52.68%, the fee-bearing reinvested SPY 95% control returned +29.91%, and full-exposure fee-free SPY context returned +31.54%.
Those original rules and receipts remain unchanged, and the new native comparison reproduced 60.31%. The broader search's training winner ranks income/assets under a P/E ceiling and uses risk parity, so its 36.19% later result is a different portfolio. All of these dates are now known research history.
Each line starts at $10,000. The chart keeps the last native observation for each New York date; tables report maximum drawdown from the complete native path. Initial purchase fees remain in the return.
The unchanged training leader on 2025 to September 2026: drawdowns
Value (%)
- Training leader: income/assets with risk parity 2026-09-30, 16:00:00 EDT
- -12.47%
- Unchanged cash/operating-profit portfolio 2026-09-30, 16:00:00 EDT
- -8.02%
- Cheap stocks only 2026-09-30, 16:00:00 EDT
- -10.51%
- SPY: reinvested dividends, initially 95% stocks 2026-09-30, 16:00:00 EDT
- -1.89%
Inspect the latest loss paths
Negative values show the loss from the running high of the displayed daily NAV observations. Sampling can make the chart extreme slightly smaller than the native maximum in the tables.
The earlier trades explain why low prices can mislead
In the earlier 2016 to 2021 research loop, the cheap-only order tape included airline, department-store and energy exposure entering the 2020 decline. Average-cost partial-sale returns after modeled fees included Macy's at-67.77%, Devon Energy at-79.94%, United Airlines at-51.78% and American Airlines at-52.90%. These examples belong to that earlier tape, not to a newly traced order history for every broad-search variant.
A profitability-filtered earlier portfolio still bought Nektar and recorded a-43.09% partial-sale loss from October to December 2018. Nektar's 2018 revenue included $1.06 billion of collaboration license revenue, and the company received a real $1 billion cash payment. The example motivates checking whether a profit measure persists; it does not show that the revenue lacked cash.
The broader cash safeguard slightly weakened the new training leader. Cash-backed earnings and recurring earnings answer different questions. The cash/operating-profit variant's lack of Nektar in the earlier full order tape does not isolate whether a filter rejected it or another stock outranked it.
The rules and comparison conventions
Read exact rules, costs and comparison conventions
Scroll sideways to compare all columns. Select a heading to sort.
| Eligible stocks | Historical S&P 500 with explicit membership filter; minimum $1 billion market cap; one share class per issuer; exclude missing capitalization |
| Valuation | Positive trailing P/E in every family; family-specific sales, book, cash-flow or enterprise-value rankings |
| Training leader | Rank highest available net income/assets; P/E below 25; up to ten stocks;90-calendar-day refresh; native risk parity |
| Unchanged incumbent | Rank lowest positive P/E; net income/assets above 5%; positive free cash flow and operating income; up to ten equal weights;90-day refresh |
| Stock allocation | 95% target exposure, rebalanced; SPY controls begin at 95% and then drift |
| Risk parity | 63-period native history; minimum 30 observations; covariance shrinkage 0.5; cold names can use equal-weight fallback |
| Execution | Native Day backtests; open/close evaluations; fractional shares; $10,000 initial capital;0.05% of stock notional per fill |
| Dividends | Stocks credit cash; refresh can redeploy it. Primary SPY control reinvests. Cash SPY leaves dividends idle. |
| Corporate-action workaround | Exclude all six Brown-Forman share-class aliases uniformly because an earlier 2016 split was absent from the trade accounting |
| Selection | Highest earlier-period total return; lower drawdown then key break ties; no added drawdown hurdle |
| Costs and coverage | Modeled fees; no separate taxes, spreads or market-impact model. Corporate actions were not independently exhaustively audited. |
No matching rows. Clear the filter to see all records.
Financial ratios use available statement observations, not a guarantee of recurring annual profit or ROIC. Missing required observations can exclude candidates; stale statements can persist.
What Aurora and the native engine did
Read what Aurora and the native engine did
Aurora proposed the economic menu, read the actual equal, earnings-yield and risk-parity drafts, and explained the differences between cheapness, quality and screening order. Its first example was an unchanged clone, so it was excluded from tested variant counts. Unsupported composite-rank, ROIC and buyback ideas were not replaced with invented formulas.
Codex translated supported ideas into exact rules, used MCP to validate and save inactive drafts, and ran the explicit native comparisons. NexusTrade's trading engine calculated the returns, fees and histories. The same frozen shortlist fed every retrospective training selection. Native histories supplied the balance and drawdown charts.
The full research retains all 60 initial configurations, eight breadth/refresh cells, four safeguard cells, fold training rankings, test controls and complete paginated histories. Actual portfolios remain inactive.
No comments yet.