Portfolio Lab

Fundamental backtest: what metric-based stock strategies would really have delivered

In depth11 min read · updated 11 November 2026 · 14 sources

A fundamental backtest checks what a rule built on metrics such as P/E, return on equity or leverage would have delivered in the past: at each date, the stocks that meet all criteria enter the portfolio. That sounds simple. But three mistakes quickly make such tests look too good: figures that were not yet known at the time, companies that no longer exist today, and too many criteria tried out. This article shows how to spot and avoid them.

From screener to backtest

A stock screener filters today: which stocks have a P/E below 15, a return on equity above 15% and little debt? A fundamental backtest asks the same question at every past date, buys the hits, holds them until the next date and compares the result with an index. A snapshot thus becomes a time series with return, volatility and maximum drawdown.

A rule has four parts: the universe (say, all US stocks with a market cap of at least one billion dollars), the criteria (metric, comparison, threshold), the weighting (equal or by market cap) and the rhythm (monthly, quarterly, annually). On top comes an answer to the question of what happens to the money when too few stocks meet the criteria. Many tests skip this case, although it occurs regularly with strict rules.

Unlike a backtest with prices alone, fundamental data arrive with a delay, only once a year or quarter, and they are sometimes changed after the fact. This is exactly where the typical mistakes come from.

Look-ahead bias: figures that did not exist yet

The most common mistake is look-ahead bias: at a date, the backtest uses figures that were only published later. A fiscal year ends on December 31, but the annual report only arrives weeks later. In the US, the annual report on Form 10-K is due 60, 75 or 90 days after the end of the fiscal year, depending on the size of the company (SEC, Financial Reporting Manual 2026). Anyone who assigns the figures to the reporting date instead of the publication date trades in the backtest with knowledge nobody had.

FilerDeadlineCalculated deadline (fiscal year to 12/31/2025)First date with the figures (month-end)
Large accelerated filer60 daysMarch 1, 2026March 31, 2026
Accelerated filer75 daysMarch 16, 2026March 31, 2026
All other companies90 daysMarch 31, 2026April 30, 2026

A report filed exactly at a month-end only counts at the following month-end, because it may have arrived after the close; that is why the last row shows April 30. An example shows why this matters: a company reports a strong jump in earnings on February 28, and the stock rises 8% the next day. A backtest that already knows the figures as of December 31 buys at the end of December and pockets the 8%. No investor could have done that.

A second form hides in databases that only store the current state of the figures. If a company later revises its earnings down, only the revised value remains in the data set, and the backtest avoids the stock before anyone knew about the problem. The remedy is point-in-time data: per report, the figures as they were published at the time, with the publication date. A generous gap is also common in research: Fama and French form their portfolios at mid-year with the prior year's financial statements so that the figures are surely known (Fama and French 1993).

Survivorship bias: today's universe

The second big mistake is survivorship bias. Anyone who takes the stocks listed today for a test since 2012 leaves out all those that were taken over, went bankrupt or delisted since then. The bankruptcies in particular are then missing along with their last, large losses. Shumway (1997) showed for the main US price database that for stocks that disappeared from the exchange for negative reasons, the last return is mostly missing and that these missing returns are large. Brown et al. (1992) already described how strongly survivorship bias distorts performance studies.

This leads to two rules for your own test. First: compare a metric-based strategy not only with an index but also with the universe it selects from, equal-weighted and with the same flaw. Second: be especially skeptical of rules that buy cheap, small or highly indebted stocks, because most of the failures are missing for them.

Many criteria, empty portfolio

Every additional criterion filters out stocks. If the criteria were independent of each other and each was met by half of the stocks, only half would remain after each criterion:

stocks in the universe
share that meets a single criterion
Number of criteria
Expected hits with 2,000 stocks and independent criteria

Calculation 2,000 × p^k; made-up universe. Real criteria are related: a low P/E and a high dividend yield often hit the same stocks, high growth and a low P/E rarely.

With five criteria that each half of the stocks meets, about 62 of 2,000 stocks remain, with ten just two. Strict thresholds like Benjamin Graham's rules for defensive investors (Graham 2003) empty the portfolio entirely in some years. A backtest then has to decide: does the money sit idle? Is it spread over the few hits so that a single stock carries the whole portfolio? Both distort the result.

A clean solution is a minimum number m with a fallback: with k hits below m, each stock gets the share it would have with m stocks, that is 1/m; the rest sits in the fallback, such as the money market or an index fund. With three hits and m = 10, 30% is in the stocks and 70% in the fallback, and with no hits everything. Diversification is preserved, and the result openly shows how often the rule found nothing.

Many criteria have a second catch: every metric and every threshold is a degree of freedom. Trying ten criteria with three thresholds each means testing tens of thousands of rules and almost certainly finding one that looked good by chance (backtest overfitting). Because of the many factors tested, Harvey et al. (2016) consider a t-statistic above 3.0 necessary for a new factor instead of the usual 1.96.

What research shows about metric-based strategies

That metrics can explain returns is well documented. Basu (1977) already found that stocks with low P/E ratios returned more than their risk suggested. According to Fama and French (1992), size and book-to-market together capture the return differences associated with beta, size, leverage, book value and P/E. Lakonishok et al. (1994) attribute the edge of cheap stocks to the behavior of typical investors, who extrapolate past growth too far, not to higher risk; together with size, such effects are behind the Fama-French factors.

Quality matters too: among stocks with a high book-to-market ratio, Piotroski (2000) separated winners from losers with a simple score of nine balance sheet signals, especially among small, little-followed firms. Novy-Marx (2013) showed that gross profitability predicts returns about as well as valuation, and according to Asness et al. (2019) profitable, growing and safe companies earn higher risk-adjusted returns.

The counter-evidence is sobering. Hou et al. (2020) replicated 452 anomalies: 65% did not reach a t-statistic of 1.96, with the stricter multiple-testing hurdle of 2.78 it was 82%, and even the confirmed ones were smaller than originally reported. McLean and Pontiff (2016) studied 97 published return predictors: outside the original sample their portfolios returned 26% less, after publication 58% less. A backtest that carries a known rule forward to today therefore mainly shows the past.

Advanced: what a point-in-time test also has to consider

  • Fiscal years: many companies do not close their books in December. A test with a fixed rhythm therefore compares figures of different ages; an age limit (such as 18 months after the reporting date) throws out stale reports.
  • Metrics with price: P/E or dividend yield change with every price, even without a new report. Monthly rebalancing thus often reshuffles the same companies and creates turnover and costs.
  • Market cap across splits: prices in databases are usually split-adjusted, share counts from old reports are not. Without conversion, market cap jumps by the split ratio.
  • Missing values: if a figure is missing, the criterion must not silently count as met. Filling missing values with the average smuggles information from other companies into the selection.
  • Trade timing: signal and purchase at the same closing price are a simplification. For metrics from annual reports the error is small, for criteria such as momentum larger.
  • Comparison: an index like the S&P 500 is weighted by market cap, whereas an equal-weighted rule already tilts toward smaller companies. The fair comparison is the equal-weighted universe.

How Portfolio Lab calculates

The fundamental backtest uses the 10-K annual reports of all US companies listed today from the SEC, optionally also the 10-Q quarterly reports with the figures of the last four quarters, per report with the values as they stood in it and the filing date. At each month-end, only what was filed before and lies no more than 18 months back counts. You choose up to ten criteria from 22 metrics, the rhythm, the weighting, a maximum and the minimum number with a fallback (money market, S&P 500 or universe). The result stands next to the S&P 500 and the equal-weighted universe, plus, per date, how many stocks the rule found and how much each criterion filtered out. Survivorship bias remains, and the page says so clearly; since November 2026 the data keeps companies that drop out, so the flaw no longer grows. Details are in the methodology.

Frequently asked questions

What is a fundamental backtest?

A test that applies a rule built on metrics such as P/E, return on equity or leverage at many past dates: each time it buys the stocks that meet all criteria, holds them until the next date and compares the result with an index and the universe.

How do I avoid look-ahead bias with financial metrics?

By using each figure only from the day it was published and taking the values as they stood in the report at the time, not today's revised state. Without filing dates, a generous gap helps, such as using prior-year statements only from mid-year.

How many criteria make sense?

As few as the idea needs. Each criterion roughly halves the universe if half of the stocks meet it, and each threshold is another degree of freedom for overfitting. Three to five criteria with round thresholds set in advance are easier to believe than ten finely tuned ones.

What happens if no stock meets all criteria?

The rule has to define that. A minimum number makes sense: with fewer hits, each stock gets the share it would have at the minimum number, and the rest sits in the money market or an index. With no hits, everything sits there.

Why do value strategies often look too good in backtests?

Because cheap, small and highly indebted stocks go bankrupt more often. If these failures are missing from the data set, as with a universe of stocks listed today, precisely the strategy's worst positions are missing.

Literature

  1. Asness, Clifford S.; Frazzini, Andrea; Pedersen, Lasse Heje (2019): Quality minus junk. Review of Accounting Studies 24 (1), pp. 34–112. doi:10.1007/s11142-018-9470-2
  2. Basu, S. (1977): Investment Performance of Common Stocks in Relation to Their Price-Earnings Ratios: A Test of the Efficient Market Hypothesis. The Journal of Finance 32 (3), pp. 663–682. doi:10.1111/j.1540-6261.1977.tb01979.x
  3. Brown, Stephen J.; Goetzmann, William; Ibbotson, Roger G.; Ross, Stephen A. (1992): Survivorship Bias in Performance Studies. Review of Financial Studies 5 (4), pp. 553–580. doi:10.1093/rfs/5.4.553
  4. Fama, Eugene F.; French, Kenneth R. (1992): The Cross-Section of Expected Stock Returns. The Journal of Finance 47 (2), pp. 427–465. doi:10.1111/j.1540-6261.1992.tb04398.x
  5. Fama, Eugene F.; French, Kenneth R. (1993): Common risk factors in the returns on stocks and bonds. Journal of Financial Economics 33 (1), pp. 3–56. doi:10.1016/0304-405x(93)90023-5
  6. Graham, Benjamin (2003): The Intelligent Investor. HarperBusiness Essentials, revised edition with commentary by Jason Zweig (first published 1949).
  7. Harvey, Campbell R.; Liu, Yan; Zhu, Heqing (2016): … and the Cross-Section of Expected Returns. Review of Financial Studies 29 (1), pp. 5–68. doi:10.1093/rfs/hhv059
  8. Hou, Kewei; Xue, Chen; Zhang, Lu (2020): Replicating Anomalies. The Review of Financial Studies 33 (5), pp. 2019–2133. doi:10.1093/rfs/hhy131
  9. Lakonishok, Josef; Shleifer, Andrei; Vishny, Robert W. (1994): Contrarian Investment, Extrapolation, and Risk. The Journal of Finance 49 (5), pp. 1541–1578. doi:10.1111/j.1540-6261.1994.tb04772.x
  10. McLean, R. David; Pontiff, Jeffrey (2016): Does Academic Research Destroy Stock Return Predictability?. The Journal of Finance 71 (1), pp. 5–32. doi:10.1111/jofi.12365
  11. Novy-Marx, Robert (2013): The other side of value: The gross profitability premium. Journal of Financial Economics 108 (1), pp. 1–28. doi:10.1016/j.jfineco.2013.01.003
  12. Piotroski, Joseph D. (2000): Value Investing: The Use of Historical Financial Statement Information to Separate Winners from Losers. Journal of Accounting Research 38 (Supplement), pp. 1–41. doi:10.2307/2672906
  13. Shumway, Tyler (1997): The Delisting Bias in CRSP Data. The Journal of Finance 52 (1), pp. 327–340. doi:10.1111/j.1540-6261.1997.tb03818.x
  14. U.S. Securities and Exchange Commission (2026): Financial Reporting Manual, Topic 1: Registrant's Financial Statements (Section 1330, filing deadlines). SEC, Division of Corporation Finance; accessed November 7, 2026. Source

Full bibliography

This article explains a method; it is not investment or tax advice. Worked examples are simplified and rounded. All terms in the glossary.