Does the value+quality score actually beat the market? — backtest writeup

2017-2024 (8 years) · top-20 equal-weight · annual rebalance

⚠ Read this before the numbers below: this backtest has a real survivorship-bias problem that is not fully solved (details below). Treat the outperformance margin as a rough, optimistic-leaning signal — not proof the strategy works going forward.

Result

	Total return	CAGR
Strategy (top-20)	+1096.0%	+36.4%
Universe, equal-weight	+171.0%	+13.3%
VTI (real fund, reference)	+132.9%	+11.1%

Average annual excess over the universe: +30.1pp · beat the universe in 5/8 years.

Year by year

Rebalanced	Universe	Strategy	Universe EW	Excess	VTI
2017	1,774	+35.8%	+23.8%	+12.0pp	+14.9%
2018	1,846	-11.6%	-0.0%	-11.6pp	+2.8%
2019	1,933	-13.5%	-5.8%	-7.7pp	+3.2%
2020	2,011	+157.9%	+84.2%	+73.7pp	+44.0%
2021	2,132	+54.3%	-10.9%	+65.2pp	-9.1%
2022	2,274	+11.3%	+5.0%	+6.3pp	+6.3%
2023	2,350	+21.5%	+23.0%	-1.5pp	+26.1%
2024	2,402	+114.1%	+9.8%	+104.3pp	+8.8%

Methodology

Top-20 equal-weight, value+quality factor score (same scoring.py used live), annual rebalance, point-in-time financials (only fiscal years whose 10-K was actually SEC-filed before the rebalance date), 0.15% round-trip cost, pre-tax dividends added.

Why two benchmarks

The 'universe' benchmark shares the exact same survivorship bias as the strategy (it's the same distorted universe, just unselected) — so the *excess* over it is more trustworthy than either absolute number. VTI (Vanguard Total Stock Market ETF) is a real fund with no survivorship bias, included as an honest floor reference for the absolute-return side, which the internal numbers cannot be.

Honest limitations

Survivorship bias: the universe is today's ~3,000-largest SEC filers. Companies that went bankrupt or were delisted between 2017 and 2024 are not in today's list and could not be included — this almost certainly inflates both the strategy's and the universe benchmark's absolute returns, likely by a meaningful amount. The official Russell 3000 holdings history (which would fix this) is blocked to automated downloads.
Share count is held fixed at today's value for every historical year — buybacks and share issuance are not reflected, unlike price, which uses the real historical value.
Dividends are added pre-tax, at the fiscal year used for stock selection — not modeled as precisely-timed cash flows.
Prices are sampled weekly, not daily.
A rough filter drops any stock-period with a single-week price move over 3x, to catch cases where a reused ticker (bankruptcy reorganization, reverse split) splices two unrelated companies' price histories together — found one such case (CHRD/Whiting Petroleum) during validation. This is a blunt instrument, not a verified per-company check.

Past performance does not guarantee future returns. This page exists so the survivorship-bias problem is visible, not buried — if the official Russell 3000 historical holdings become accessible, this backtest will be rebuilt properly and this page updated.

← See the full screener