Transparent Validation

Methodology

Phoenix uses multiple portfolio models designed for different investing timeframes. Some focus on shorter-term opportunities, while others look for stronger longer-term setups. Each model is tested on historical market periods it has never seen before to help measure how well it generalizes over time.

Results include periods where the model underperformed the market.

2016-2026 historical data range used in the current training set

503 symbols S&P 500-sized equity universe represented in validation data

~1.2M rows labeled historical examples per horizon after cleaning

85/15 split chronological train/validation split to reduce look-ahead bias

How The Portfolio Is Built

Three independent horizons Phoenix trains and scores separate models for short, medium, and long holding windows. Each is judged on forward returns for that same horizon.

Out-of-sample validation Models are trained on earlier dates and tested on later dates that were not used in training. Current validation windows run from 2024 into 2026, depending on horizon length.

Current production assembly The system updates daily using the latest market data, builds portfolio candidates across multiple time horizons, and combines the strongest signals into one final portfolio list.

Risk and liquidity review Medium and long horizon selections include a post-model screen to avoid the weakest risk and liquidity profiles in the candidate pool.

Assumptions

Position sizing Validation uses equal-weight Top-5 baskets for each date.

Rebalancing Each validation date forms a fresh Top-5 basket for the tested horizon.

Fees and slippage Returns shown here do not deduct commissions, spreads, taxes, or slippage.

Purpose These results measure portfolio simulation quality, not future return guarantees.

Historical Validation Summary

Horizon	Data range	Validation range	Rows	Symbols	Days tested	Avg overlay top-5 return	Hit rate	Worst period	Best period	Approx. annualized
10-day	2016-04-25 to 2026-02-24	2024-08-27 to 2026-02-24	1,209,625	503	374	+0.83%	52.14%	-14.67%	+23.31%	+23.2%
30-day	2016-03-28 to 2026-02-12	2024-08-19 to 2026-02-12	1,216,164	503	373	+3.08%	61.29%	-25.40%	+30.34%	+29.0%
90-day	2016-03-28 to 2026-01-14	2024-07-25 to 2026-01-14	1,206,104	503	370	+9.56%	63.03%	-25.45%	+65.99%	+29.1%

Medium and long horizon rows use their production overlay screens. The short horizon row uses a direct Top-5 portfolio selection. Approximate annualized figures are derived from average validation-period returns and horizon length. They are included for scale only.

Combined Portfolio Replay

Direct price-series baseline A separate combined-list replay using direct historical closes from the database, equal-weight sizing, and no exit strategy produced an average return of +16.22% across 529 historical portfolios, with an 83.93% positive-rate and a median return of +13.17%.

Construction This replay uses the merged daily combined list, enters at the nearest available close on the prediction date, exits at the nearest available close on the planned term date, and does not include fees, slippage, or taxes.

Reliability Notes

Later-date testing Validation uses later dates than training, which is more demanding than random row splitting.

Large sample Each horizon is tested across hundreds of validation dates and more than one million labeled examples.

Underperformance is retained Worst-period results are shown directly rather than removed from the summary.

Educational use Phoenix is a research and portfolio simulation tool. It is not financial advice and cannot guarantee future performance.