98 years of the market,
tested honestly.
Two questions, answered with old market data. Did simply holding the market make money? And did any of the well-known strategies that try to beat it actually do so? To keep the test fair, nothing here was searched for. Each strategy was published by someone else with its settings fixed, and it is judged only on the years after it was published, which its author couldn't have seen.
The short answer
In this data, the approach that made money most reliably was the plainest one: holding the whole market. Every one of the ten published strategies tested here made less money than that in the years after it was published. What they did do well was shrink the worst crashes.
But "holding makes money" is one country's history, not a law. Japan's market took 34 years to get back to its 1989 peak.
Did holding the market make money?
Every month since the start of the data, imagine buying and holding for a set number of years. How often were you ahead at the end?
| Held for | US S&P 500, 1928–2026 (prices only) | US S&P 500, 1988–2026 (dividends reinvested) | Japan Nikkei 225, 1965–2026 (prices only) |
|---|---|---|---|
| 1 year | Ahead 69% of the time · worst −70% | Ahead 84% · worst −43% | Ahead 65% · worst −49% |
| 5 years | Ahead 80% · worst −71% | Ahead 88% · worst −29% | Ahead 69% · worst −55% |
| 10 years | Ahead 89% · worst −65% | Ahead 93% · worst −29% | Ahead 69% · worst −63% |
| 20 years | Ahead 97% · worst −52% | Ahead 100% · worst still +4.8% a year | Ahead 73% · worst −76% |
"Prices only" leaves out dividends, which historically added a few percent a year, so those columns understate what holders got. None of these numbers subtract inflation. The typical US year (median) was about +10% on prices alone.
Did the famous "beat the market" strategies work?
Each one judged only on the years after it was published, against simply holding over exactly the same months. Cash earns Treasury-bill interest; each switch in or out costs 0.05%. Taxes are not included, which flatters every strategy except holding.
| Strategy (who published it) | Judged on | Strategy per year | Holding per year | Strategy worst drop | Holding worst drop | Verdict |
|---|---|---|---|---|---|---|
| Be in stocks only above the 10-month average (Faber, 2007) | 2007–2026 | +9.1% | +11.0% | −22% | −51% | Less money, smaller crashes |
| "Sell in May": stocks Nov–Apr only (Bouman & Jacobsen, 2002) | 2002–2026 | +7.2% | +11.5% | −30% | −51% | Less money, by more than luck |
| 12-month trend following (Moskowitz, Ooi & Pedersen, 2012) | 2012–2026 | +11.1% | +15.0% | −20% | −24% | Less money, by more than luck |
| Dual momentum: US vs. international stocks, else bonds (Antonacci, 2014) | 2014–2026 | +8.0% | +13.6% | −21% | −24% | Less money, by more than luck |
| Hold the S&P 500 only overnight (Cliff, Cooper & Gulen, 2008) | 2008–2026 | +0.3% | +11.2% | −36% | −52% | Far less money once trading costs count |
| Turn of the month: stocks only on the last day and first three days (Lakonishok & Smidt, 1988) | 1989–2026 | +5.0% | +11.4% | −19% | −55% | Less money, by more than luck |
| Stocks only on Federal Reserve announcement days (Lucca & Moench, 2015) | 2016–2026 | +1.1% | +15.1% | −7% | −34% | The effect vanished: those days now average −0.03% |
| The calmest fifth of the S&P 100 (Baker, Bradley & Wurgler, 2011) | 2011–2026 | +11.4% | +14.2% | −32% | −34% | Less money |
| The fifth closest to their 52-week high (George & Hwang, 2004) | 2009–2026 | +13.5% | +14.9% | −26% | −34% | Less money |
| Buy last week's 10 biggest losers (Jegadeesh; Lehmann, 1990) | 2009–2026 | +2.3% | +14.7% | −58% | −34% | Far less money, by more than luck |
Before they were published, two of these looked better than holding (12-month trend +11.2% vs +9.3%; dual momentum +12.1% vs +8.3%). That's the usual pattern: an edge that's found in old data shrinks once people know about it. This site's own overnight picks show the same thing. See that test →
The one idea that beat holding, and why it's still on paper
Ten published strategies, a machine-learning test and this site's own rules all made less than simply holding the market once judged fairly. One idea of ours did beat it: hold SPY during the day and switch into the site's momentum picks every night, so the money is always invested. From 2009 to 2026 it made +21.8% a year against +14.7%, and it beat holding in both halves of that period, but only if each trade costs about 0.005% or less. At 0.01% per trade it was a tie; at 0.02% it lost badly.
It trades four times a day, counts as day trading under FINRA's pattern-day-trader rule (a margin account with at least $25,000), every gain is taxed as short-term, and the idea came after looking at all of this data. So it's being tracked forward on paper from October 1, 2026. Follow it on the overnight page →
Method and limits
- Data: Yahoo Finance. S&P 500 prices from December 1927, S&P 500 total return (dividends reinvested) from 1988, 13-week Treasury bills from 1960, SPY / EFA / AGG funds for dual momentum, Nikkei 225 from 1965.
- Each strategy decides at a month's last close using only data up to then, and earns the next month's return. The overnight strategy buys at each close and sells at the next open, at 0.01% per trade.
- "By more than luck" means the 95% range for the average monthly difference against holding stays below zero.
- Not included: taxes, inflation, and the effect of adding money over time. Results are nominal dollars.
- Tested October 1, 2026 (two rounds, every rule written down before running). Code and the full write-up: experiments/history_lab/ in this project.
This is historical research, not investment advice, and PontisAi is not a financial adviser. Past results do not predict future returns. For decisions about your own money, a fee-only fiduciary planner can look at your situation; the SEC's investor.gov is a free place to start.