Most of what a trend-following system does is hold the trend. The interesting question is what it does with the part of a move that is not trend: the stretch at the top, where price has run a long way from its own average and the last leg starts to look like enthusiasm rather than direction.
This one steps aside there. It holds a 3× leveraged NASDAQ-100 ETF while the trend is intact, leaves when price gets unusually far above its moving average, and buys back in stages on the way down. Over 6.7 years it returned 84.7% a year against 32.5% for simply holding the same ticker, with a worst drawdown of 33.5% against 81.7%.
The second pair of numbers is the one that matters, and the article comes back to why.
The rules
The instrument is TQQQ, daily bars, 2020-01-02 to 2026-09-11, starting from $100,000. Orders are decided on the close and filled at the next open, so no rule below can see a price that had not printed when the decision was made.
- Hold while the trend holds. The baseline position is long the instrument, whenever price is above its moving average and that average is itself rising. The length of the average is Withheld: this is the firm's own calibration. The mechanism is described in full, and every result on this page was produced with the real value..
- Leave when price stretches too far above it. Far enough is defined either as a multiple of the average or as a number of standard deviations above it. Two profiles of one idea, and the repository ships both. The threshold is Withheld: this is the firm's own calibration. The mechanism is described in full, and every result on this page was produced with the real value.. This is a bet on reversion to the trend, not a call on a top.
- Leave on a confirmed cross below. A single close through a flat average is mostly noise, so this exit requires several consecutive closes beyond it and the next bar to open beyond it as well. How many closes is Withheld: this is the firm's own calibration. The mechanism is described in full, and every result on this page was produced with the real value..
- Ignore crossings while the average is flat. In that regime a crossing carries the least information, so the position is held through it. The stretch exit still fires: a violent run happens in a steep trend, which is exactly when leaving is worth something.
- Scale back in on the way down. After an exit, buy in steps as price falls, measured either from the average or from the running peak. Each rung deploys an equal dollar amount of the remaining cash, so a deeper fall buys more shares. The rungs are Withheld: this is the firm's own calibration. The mechanism is described in full, and every result on this page was produced with the real value.. Anything still in cash when price breaks back above the average goes in at once.
- Re-enter on the band, or on a big enough give-back. Back in when price falls to the entry band, or (optionally) once the stretch has retraced far enough from its own peak, whichever comes first.
Rule 4 and rule 5 are the two that took the longest to settle, and they solve opposite problems. The slope filter stops the system trading itself to death in a sideways market. The ladder stops it from putting the whole position back on in one guess about where the bottom is.
What it did
- CAGR
- 84.7%
- Max DD
- -33.5%
- Sharpe
- 2.22
- MAR
- 2.53
| CAGR | Max DD | Sharpe | Sortino | Volatility | MAR | |
|---|---|---|---|---|---|---|
| MA Divergence | 84.7% | -33.5% | 2.22 | 1.97 | 39.0% | 2.53 |
| Buy and hold TQQQ | 32.5% | -81.7% | 0.73 | 1.00 | 73.6% | 0.40 |
Both lines start with the same cash, trade the same bars and pay the same execution rules. The benchmark is not an index. It is this exact instrument, bought on the first day and held.
For the first two years those lines are nearly the same line, and that is the correct result rather than a disappointing one. In a straight uptrend a strategy whose baseline is hold the instrument has nothing to add, and any version of this that pulled ahead in 2020 and 2021 would be doing something other than what it says it does.
The separation is 2022. Buy-and-hold fell 81.7% peak to trough and spent the next two years getting back to where it started. The strategy was in cash for much of it, lost 13.33% on the year, and compounded from 2023 onward off a base it had never given back.
Beating an 81.7% drawdown with a 33.5% one is a more meaningful result than beating 32.5% a year with 84.7%. The return figure describes a backtest. The drawdown figure describes whether a person is still holding the position when the recovery arrives, and an 80% loss on a leveraged ETF is the kind of number that ends the experiment before the recovery is relevant.
A third of the account is still a bad year by any ordinary standard. It is only respectable relative to the instrument.
Year by year
| Year | J | F | M | A | M | J | J | A | S | O | N | D | Year |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2020 | 0 | -10 | 21 | 10 | 11 | 8 | 0 | 0 | 0 | 32 | 3 | 108% | |
| 2021 | 0 | 0 | 11 | 21 | 13 | 19 | 7 | 0 | 0 | 29 | 11 | 0 | 180% |
| 2022 | -5 | 0 | 7 | 0 | 0 | -5 | 14 | -7 | -12 | 6 | 7 | -16 | -13% |
| 2023 | 17 | -2 | 17 | -0 | 15 | 10 | 0 | 0 | -3 | -8 | 34 | 16 | 138% |
| 2024 | 13 | 0 | 0 | 1 | 19 | 19 | 5 | -2 | 6 | -4 | 15 | -1 | 101% |
| 2025 | 5 | -9 | -7 | 0 | 1 | 19 | 6 | 1 | 16 | 17 | 16 | -3 | 73% |
| 2026 | 2 | -8 | -16 | 52 | 19 | 0 | -2 | 11 | -1 | 57% |
Monthly return in percent. Colour saturates at 40%.
Six up years and one down year, and the down year is the informative row rather than the worst one. 2022 is the only period in this sample where the system had to defend rather than compound, and it ended it down 13.33% while the instrument it tracks fell 79.68%. Everything else in the table is a bull market being a bull market.
That is also the sample's central weakness: 6.7 years containing exactly one bear market is one observation of the only thing this design is really for.
The trades
16 completed round trips in 6.7 years, from 36 fills. This is not a high-turnover system.
| Win rate | 81.2% (13W / 3L) | Profit factor | 21.94 |
| Longest win streak | 5 | Longest losing streak | 1 |
| Median hold | 47.5 days | Longest hold | 401 days |
| Average | Median | Worst | Best | |
|---|---|---|---|---|
| Return per round trip | +43.09% | +39.96% | -18.63% | +194.2% |
| Winners | +55.7% | +43.75% | +8.87% | +194.2% |
| Losers | -11.54% | -11.43% | -4.56% | -18.63% |
A profit factor near 22 is not evidence of a miraculous edge, and reading it that way is the main mistake available here. It is what arithmetic does when a strategy takes small capped losses and occasionally rides a leveraged instrument through a long trend. The shape is the part worth having: every loss in this sample falls between -4.56% and -18.63%, a tight band, while the winners are open-ended and the best ran 401 days.
It is also 16 observations. A win rate computed on sixteen trades carries error bars wide enough to drive through, and nothing in that table should be treated as a stable property of the system.
The shape of the returns
| Period | Count | Worst | Median | Average | Best | % positive |
|---|---|---|---|---|---|---|
| Daily | 1,052 | -20.31% | +0.48% | +0.44% | +13.52% | 58% |
| Weekly | 350 | -12.85% | 0% | +1.02% | +28.86% | 40% |
| Monthly | 81 | -16.01% | +3.15% | +6.04% | +46.98% | 58% |
| Yearly | 7 | -13.33% | +101.32% | +91.97% | +179.62% | 86% |
The weekly row is the honest one. Only 40% of weeks are positive and the median week is exactly zero, because the strategy is frequently in cash. Returns arrive in concentrated bursts separated by long stretches of nothing happening, which is uncomfortable to sit through in real time no matter what the annual figures say.
The distribution also gets prettier the further you zoom out: 86% of years are positive against 58% of days. That is true of every volatile strategy and is evidence of nothing.
What I do not know
- The parameters were chosen by looking at backtests. They are worth exactly what they do out of sample and no more. On a system like this, adjacent parameter values differ by tens of CAGR points. That spread is noise, and picking the top of it is the standard way to fool yourself.
- One regime and a bit. 2020–2026 was, with one interruption, an extraordinary run for leveraged NASDAQ exposure. A strategy whose baseline is hold TQQQ inherits all of that. The question worth asking is not whether it beat cash; it is what it does in a decade that looks like 2000–2010, which is not in this sample.
- The right comparison is buy-and-hold, not zero. It is in every table above for that reason. Against cash almost anything looks clever.
- Costs are modelled but light. Whole-share fills at the next open, no commission, no borrowing cost, and no slippage assumption beyond using the open price.
- Leveraged ETFs decay. TQQQ resets its leverage daily, so its long-run behaviour depends on the path and not only the destination. A backtest captures that only as far as the price history does, and the price history here is one path.
Before believing any figure on this page, run the split: a parameter worth keeping sits near the top of both halves of the period, not at the full-period maximum. The engine ships the tools to do it: grid sweeps, walk-forward efficiency, Deflated Sharpe, CSCV/PBO and a robustness gate. A companion article works through what happens when six different rules are used to pick one configuration out of the same sweep.