In Part 2, I ran Shannon’s Demon and eight different trend-following systems against 20 years of real Nasdaq-100 data, and every single one of them lost to the most boring strategy imaginable: buying every month and never selling, no matter what the chart looked like.
At first I assumed this was mostly a “cost basis” problem — trend systems buy back in at a slightly worse average price than someone who never sold at all, because trend signals are always a little late. That’s a real effect, but when I actually ran the numbers, it turned out to be a minor character in this story. The real villain (or hero, depending on which side of it you’re on) is something much more specific and much more brutal.
Missing the Best Days in the Stock Market: Why 10 Days Out of 5,000 Matter Most
Out of ~5,000 Trading Days, Only a Handful Actually Mattered
Over the ~20-year window, QQQ traded on roughly 5,000 separate days. I ran a simple experiment: take the lump-sum buy-and-hold result (18.14x), then recalculate what happens if you strip out just the single best-performing days — the biggest single-day percentage gains — one at a time.
Read that again. Missing just 10 days out of roughly 5,000 — 0.2% of all trading days — cuts your final return by more than half. Miss 30 days, and you’ve given up over 83% of the entire 20-year gain. This is one of the more genuinely startling things you can do with a spreadsheet and 20 years of price data, and it’s not a QQQ-specific quirk — the same pattern shows up in basically every study of long-run equity index returns.
Why This Wrecks Every Trend-Following Strategy
Here’s the part that connects directly back to Part 2. Those monster up-days aren’t scattered randomly across the calendar. They cluster tightly around market crashes — the days immediately following a panic-driven crater, when fear turns into a violent, disorganized scramble back in. Some of the biggest single-day percentage gains in QQQ’s history happened within days of the 2008 financial crisis lows and the March 2020 COVID crash.
And that is exactly the moment a trend-following system is guaranteed to be sitting in cash. By construction, these systems exit during the crash (that’s the whole point of a stop-loss) and only re-enter once the trend is “confirmed” — which, mathematically, cannot happen until the price has already clawed back a meaningful chunk of the crash. The system is, by design, engineered to be on the sidelines for the exact days that matter the most.
I even tried to patch this: what if, while sitting in cash, the system watches for a single explosive up-day (say, +5% in one session) and jumps back in immediately, no confirmation needed, no waiting? It helped — a 200-day moving average exit strategy improved from 5.12x to about 7.46x with a well-tuned trigger. But it never closed the gap with simply never having left in the first place (7.68x, no rules at all), and in several versions it actually made drawdowns worse — because a violent one-day spike doesn’t reliably distinguish a real bottom from a “dead cat bounce” in the middle of a continuing crash. Roughly half the biggest up-days in the data were the start of real recoveries; the other half were followed by another leg down. In real time, with no hindsight, there is no clean way to tell those two apart.
This isn’t just a quirk of this particular backtest. J.P. Morgan Asset Management’s own research on the S&P 500 found something almost identical: missing just the market’s 10 best days over a 20-year window (2005–2024) cut an investor’s return from 10.6% annualized down to roughly 6.4% — turning a $71,750 outcome into barely half that. The fact that two completely independent analyses, on two different indexes, land on the same conclusion is a strong signal this isn’t a statistical fluke.
(Source: J.P. Morgan Private Bank)
So What’s Actually Going On Here?
Put the whole series together and a clear picture emerges:
- Shannon’s Demon is real math, applied to the wrong asset. It genuinely creates return from rebalancing — but only on something that chops sideways. Feed it a two-decade bull market and it just repeatedly sells your winners and never lets them run.
- Trend-following stop-losses genuinely do what they promise — they meaningfully cut max drawdown, in some tests by more than half. That’s a real, honest benefit if your priority is sleeping at night during a crash.
- But that safety has a specific, quantifiable price: trend systems are structurally guaranteed to miss the handful of explosive recovery days that drive a disproportionate share of all long-term returns — and those few days are worth more than years of “smart” trading around them.
- Simple, disciplined, unglamorous dollar-cost averaging — buying on a schedule, doing nothing clever, riding out every crash and every recovery — beat every single “smarter” alternative tested, on both a strongly bullish 20-year window and on the risk-adjusted table too.
None of this means risk management is pointless — if a 39% drawdown would genuinely make you sell everything at the worst possible moment, a rule that caps your drawdown at 25% might save you from yourself, and that’s worth something real. But it’s worth being honest about the price tag: in this data, every point of drawdown protection was bought with several points of long-run return, and none of the tested rules found a way around that trade-off.
The Practical Takeaway
You don’t need to be Claude Shannon, Ed Thorp, or a hedge fund quant to act on any of this. The entire, unglamorous conclusion of three articles’ worth of backtesting is:
- Automate your contributions. A fixed amount, on a fixed schedule, into a broad, structurally self-cleaning index (something like a total-market or Nasdaq-100 fund, where losers get dropped and winners get added automatically) — and then leave it alone.
- Don’t try to dodge the crash. The instinct to get out during a downturn feels protective, but the data says the real danger isn’t the crash itself — it’s the handful of ferocious recovery days immediately afterward that you’re statistically almost guaranteed to miss if you’re waiting for the “all clear” signal.
- Save the clever stuff for the things you can actually verify in real time — rebalancing across genuinely uncorrelated assets, tax-loss harvesting, position sizing on individual high-conviction bets. Just don’t expect it to outperform simply staying invested in a long-term-growth index.
Claude Shannon and Ed Thorp built careers out of finding tiny, provable, repeatable mathematical edges and exploiting them mercilessly. It turns out that after two decades of real market data and nine different “smart” strategies, the biggest edge available to an ordinary investor isn’t some clever formula at all. It’s just staying in your seat.
This concludes our three-part series: from the elegant theory of Shannon’s Demon, through a reality check against 20 years of Nasdaq-100 data, to the simple math of why time in the market beats timing the market. If you backtest your own favorite “smart money” trick, I’d genuinely love to hear what you find.