The Same Plan, Eleven Different Starting Years
Every backtest reports one number, and it is easy to read that number as what the strategy returns. It is not. It is what the strategy returned over the window you happened to pick. Here is the same plan run over eleven of them.
#The test
One plan: $500 a month into SPY for ten years, dividends reinvested. Every run contributes exactly $60,000 and lasts exactly ten years. The only thing that changes is the year it began.
| Started | Invested | Final value | Ann. return |
|---|---|---|---|
| 1995 to 2005 | $60,000 | $83,683 | +6.58% |
| 1997 to 2007 | $60,000 | $81,857 | +6.15% |
| 1999 to 2009worst | $60,000 | $49,601 | -3.89% |
| 2001 to 2011 | $60,000 | $73,855 | +4.14% |
| 2003 to 2013 | $60,000 | $80,674 | +5.86% |
| 2005 to 2015 | $60,000 | $105,888 | +11.11% |
| 2007 to 2017 | $60,000 | $106,677 | +11.25% |
| 2009 to 2019 | $60,000 | $106,887 | +11.29% |
| 2011 to 2021 | $60,000 | $126,613 | +14.50% |
| 2013 to 2023 | $60,000 | $101,844 | +10.37% |
| 2015 to 2025best | $60,000 | $127,340 | +14.61% |
#The spread
The worst window, starting in 1999, finished at $49,601. That is an annualized return of negative 3.89 percent, and it is the only run of the eleven that ended below what was contributed. Ten years of disciplined investing, $60,000 in, and about $10,000 destroyed.
The best window, starting in 2015, finished at $127,340, an annualized 14.61 percent. More than two and a half times the ending balance of the worst one.
The full spread in annualized return is 18.51 percentage points, from negative 3.89 to positive 14.61. Nothing about the plan differed. The investor made the same decision, in the same fund, for the same length of time, and contributed the same money.
#What the pattern shows, and what it does not
There is a clear shape here. The windows that began in the late 1990s and early 2000s did poorly; the ones that began after 2005 did well. That is not a trend you can extrapolate. It is the dot-com crash and the financial crisis landing inside some windows and not others.
Notice that 1999 to 2009 is far worse than its neighbors on both sides. A plan starting in 1997 returned 6.15 percent and one starting in 2001 returned 4.14 percent, while the one between them lost money. Shifting the start by two years in either direction changes the answer completely.
The reason is covered in sequence of returns risk: the 1999 window put its heaviest accumulated balance directly into the 2008 crash with no time left to recover, while the 2001 window bought through the dot-com bottom first and had a larger cushion by the time 2008 arrived.
#Why this is not an argument against investing
Ten of the eleven windows finished ahead, and nine of them returned more than five percent annualized. The median run returned 10.37 percent. The bad outcome is real, and it is one out of eleven.
It is also worth noting what happened to that unlucky 1999 investor afterward. Their plan looked like a failure at the end of 2009. Anyone who continued contributing through the 2010s did well, because the following decade was one of the strongest on record. Ten years is a long time to an investor and a short window to a market.
- A single bad decade is survivable if it happens early, when little is invested and decades of contributing remain.
- The same decade arriving just before retirement is a materially different problem, because there is no time left to contribute through it.
- The spread narrows as the window lengthens. Run twenty-year windows instead of ten and the range of outcomes compresses substantially.
#What to do with this
The practical use is calibration. When you run a comparison on this site and see an annualized return, the honest way to read it is as one draw from a distribution roughly this wide, not as the number the strategy produces.
So run more than one window. If a conclusion holds across a start in 1999, 2007, and 2015, it is probably about the strategy. If it only appears in the window you happened to choose first, it is about the window. That is the discipline described in why backtests can mislead you, and this page is the clearest demonstration of why it matters.
It is also an argument for the thing this spread cannot touch: your contribution rate. You do not control which decade you get. You do control how much goes in.