We ran an experiment: backtest the same strategies using today’s SP500 constituent list vs. the historically accurate point-in-time list. The gap is staggering — momentum returns were overstated by 102%, equal-weight rebalancing by 42%.
The Problem: What Is Survivorship Bias?
When you backtest an SP500 strategy, you need to know which stocks were in the index on each historical date. Most free data sources give you only the current list — the 500+ stocks that are in the index today.
The problem: the companies still in the SP500 today are the survivors. Lehman Brothers, Enron, WorldCom, Silicon Valley Bank — they’re gone. They were removed from the index after collapsing. But if you backtest using today’s list, those losers simply don’t exist in your historical data.
You’re only counting the winners. And that systematically inflates your returns.
The Experiment: 4 Strategies × 2 Groups
We compared two groups over a 10-year period (2016–2026):
- Group A (Biased): Uses the July 2026 SP500 constituent list — the 503 stocks in the index today. This is what most free backtesting tools do.
- Group B (Point-in-Time): On each rebalancing date, queries the actual constituents that were in the index on that specific date, sourced from Wikipedia’s audited change log (899 verified records since 1957).
Both groups use identical strategy logic, the same rebalancing frequency (monthly), and the same price data. The only variable is which constituent list is used.
The Results
| Strategy | A — Biased | B — Point-in-Time | Bias / Year | Overstatement |
|---|---|---|---|---|
| Momentum Top 20% | +21.5%/yr | +10.6%/yr | +10.85% | 102% |
| Equal-Weight Rebalance | +17.4%/yr | +12.2%/yr | +5.14% | 42% |
| Equal-Weight Hold | +19.8%/yr | +15.0%/yr | +4.79% | 32% |
| Low Volatility | +8.1%/yr | +7.0%/yr | +1.14% | 16% |
A strategy that looks like it returns 21.5% per year actually returns 10.6% — nearly cut in half. That’s the difference between “this strategy is amazing” and “this strategy barely beats the market.”
Why Momentum Suffers Most: Double Bias
Momentum has the largest bias (102%) because of two effects compounding:
- Survivorship bias: The current SP500 list only contains winners. Momentum selects the strongest stocks from an already-winnowed pool — double-counting the survival filter.
- Time-travel bias: Using today’s list lets Group A “invest” in stocks that joined the SP500 years in the future. For example, ARM was added to the index in 2026 — but Group A could select it back in 2018, before it was even a constituent. You’re trading on information that didn’t exist yet.
The ranking follows a clear pattern: bias = survivorship exposure × selection amplification. Momentum (double bias) > Rebalance (monthly amplification) > Hold > Low-volatility (rarely touches delisted names).
What This Means for You
If you’re backtesting with a free SP500 list:
- Your momentum strategy’s 21% annual return is really 10% — barely above the risk-free rate.
- Your equal-weight strategy is overstated by 32–42%, enough to change the conclusion from “worth running” to “not worth the risk.”
- Even low-volatility strategies, which screen for stability, are inflated by 16%.
There is no strategy that’s immune. The only fix is using point-in-time constituent data — the exact stocks that were in the index on each historical date, including the ones that were later removed.
Stop Letting Survivorship Bias Lie to You
Get point-in-time SP500 constituent data. 899 verified records since 1957. 100% delisted-stock coverage.
Try the Survivorship Bias Lab →
How to Fix It
Stockmere provides point-in-time SP500 constituent data sourced from Wikipedia’s complete historical change log:
- 899 verified records spanning 1957 to 2026, each with
effective_from/effective_todates - 873 unique tickers (vs. only 503 in the current list) — including all delisted/renamed/merged companies
- 100% delisted-stock coverage with ticker mapping for renamed/merged companies
- Accessible via REST API or MCP for AI agents
View API Documentation → Subscribe to Point-in-Time Data →
Not investment advice. Past performance does not guarantee future results. Data compiled from Wikipedia “List of S&P 500 companies”, licensed under CC BY-SA 4.0. Not official S&P Dow Jones Indices data.
← Back to Articles