Every strategy on this site is a rule. Not "buy when it feels right" but "when condition A happens, buy a set amount; when condition B happens, sell" — repeated mechanically, one day at a time. Because they are rules, they can be replayed over historical data, which is what makes a backtest possible at all.
Except for the GDXU-only G family and the I family (TQQQ and SOXL, from 2011), all the numbers below come from running SOXL from 2010 to 2026 with a yearly reset: each January starts fresh with $10,000 and settles in December. That avoids the illusion of one lucky year carrying the whole record. You can reproduce any of them on the backtest page by selecting the strategy and pressing 📅 30-Year Report.
There are five families whose engines are completely different. The letter in front of the name tells you which one.
| Family | Core idea | Selling rule | Character |
|---|---|---|---|
| A · Split buying | Buy one tranche on each decline | Each tranche at its own target | Simple, easy to follow |
| D · Dual mode | Switch between attack and defence on market temperature | Different target per mode | Middle ground, trades often |
| B · Breakout | Buy the day price rises a set distance above the open | Everything at the next open (or at the stop that day) | Overnight hold, trades almost daily |
| G · GDXU only | Rules fitted to GDXU: breakouts and close-buy lots chosen by market regime | Next open · open three sessions later · same-day close | GDXU only, trades almost daily |
| I · Infinite buying | Laoer's V4.0: split the cash into units and buy a little at the close every day | A quarter at the star point above the average, three quarters at +15% (SOXL +20%) | For TQQQ and SOXL, orders every day |
The basic one. Capital is divided into six tranches and one tranche goes in on each day the price falls past the trigger. The reason for not buying all at once is simple: you need money left to buy lower, or your average cost never comes down.
On any day the close moves by the buy trigger versus the previous close. For A2 Standard that is −0.01%, which in practice means "buy if it closed even slightly lower." Once all six tranches are used it stops buying and waits.
Each tranche is sold separately. If the tranche bought third reaches its take-profit measured against its own purchase price, only that tranche is sold; the rest keep running. Measuring against the tranche's own cost rather than the blended average is what lets the position unwind piece by piece on every bounce, even inside a downtrend.
On top of that sits the time cut. A tranche that has not reached its target within the set number of days (10 by default) is closed anyway, at a profit or a loss. Without that single rule, a year like 2018 or 2022 ends with all six tranches frozen.
A1, A2 and A3 differ in only three things: tranche weights, take-profit and time cut. The engine is identical. Since October 2026, A2 and A3 also carry a guard signal (explained below the table).
| Strategy | Weights | Take profit | Time cut | Avg return | Avg MDD | Losing years |
|---|---|---|---|---|---|---|
| A1 Conservative | 5·10·15·20·25·25 bottom-heavy |
0.01% | 10d | 23.6% | −14.0% | 0 |
| A2 Standard | 10·15·20·25·20·10 diamond |
1.5% | 10d | 37.0% | −21.1% | 1 |
| A3 Aggressive | 16.7% each | 2.0% | 12d | 40.1% | −21.8% | 2 |
Read it this way. A1 takes profit at 0.01%, so it sells almost the moment a tranche is above water. It skims a very thin slice — and it went seventeen years without a single losing year. A3 holds out for 2%, earning more but staying stuck longer and falling deeper. Raising the take-profit raises return and drawdown together: that is the whole of the A family.
The A2/A3 guard (added October 2026). It switches on after 7 sessions in a row in which the semiconductor index (SOXX) is below its 200-day average and SOXL's 20-day volatility is above 120%. While it is on, A2 uses a 5-day time cut for tranches it buys, and A3 does not start a new ladder. A3 also sells tranches 4 and deeper at the next close whenever SOXL's 7-day volatility is more than 1.2x its 60-day volatility. It cut the deepest drawdowns, such as April 2025 (continuous 2010–2026 MDD: A2 −42% → −36%, A3 −53% → −35%), at the cost of earning less in snap-back years like 2022. A1 is unchanged.
The weakness of the A family is that it never looks at the market. Bull run or crash, it buys the same way. The D family adds a mode switch.
It reads the weekly RSI. RSI compares recent gains against recent losses: near 0 means beaten down, near 100 means overheated. Weekly rather than daily, because the daily RSI whips around enough to flip the mode every other day.
Think of it as one account that changes personality with the season. The two values you can tune are the attack tranche count and the buy baseline (how far the price must move before it buys).
The name sounds like a trigger, but it actually decides where the limit order sits. Picture placing this order before every close:
Limit price = yesterday's close × (1 + buy baseline)
It fills if today's close is at or below that price.
So a lower number is pickier and a higher number is more relaxed — the opposite of what the sign suggests, which is where people trip up.
The trade counts show it plainly. SOXL, 2021–2024, 1,005 trading days:
| Buy baseline | Meaning | Trades |
|---|---|---|
| −0.1% (D1) | down days only | 438 |
| +0.5% (D2) | skips days up more than 0.5% | 450 |
| +8% (D5) | skips only spikes (almost every day) | 538 |
Why does the relaxed setting score better? Because attack mode is already the stretch the strategy judged to be trending. Holding out for a cheaper entry inside that stretch usually means the price simply runs away and the tranches never get filled. What the D family is saying is this: being picky about entries is the defence mode's job; in attack mode, just fill the tranches.
| Strategy | Attack tranches | Buy baseline | Avg return | Avg MDD | Losing years |
|---|---|---|---|---|---|
| D1 | 6 | −0.1% (down days only) | 47.0% | −15.8% | 1 |
| D2 | 6 | +0.5% | 48.7% | −19.1% | 3 |
| D3 | 5 | +0.5% | 50.3% | −18.0% | 3 |
| D4 | 7 | +8% (almost daily) | 47.2% | −16.9% | 2 |
| D5 | 6 | +8% (almost daily) | 51.7% | −18.2% | 3 |
Bear guard on D1, D3, D4, D5 and D6 (replaced October 2026). Every current D strategy now carries a bear-market guard (D2 has been removed and stays in the table only for reference). While the SOXX 150-day average is falling, new buys are smaller, and once the account is more than 15% below its peak, new buys pause or shrink sharply. D3, D4, D5 and D6 also sell their deepest attack tranches at the next close when SOXL's 7-day volatility jumps well above its 60-day volatility. Drawdowns in crashes like 2011 and 2014 fell sharply (continuous 2010–2026 MDD: D1 −46% → −24%, D5 −49% → −26%), at the cost of earning less in big snap-back years like 2022 and 2025.
D6 (replaced again on 16 October 2026). Re-tuned without assuming any interest on idle cash. In a bear phase new buys are 0.75x; new buys pause when the account is more than 15% below its peak in a bear phase, or more than 18% below while SOXX is under its 50-day average after a recent bear phase; and attack tranches 5–6 are sold at the next close when SOXL falls more than 12% in two days or its short-term volatility spikes. Yearly reset 2010–2026: average return 54.6%, average MDD −18.3%, 3 losing years.
Against the A family the returns are clearly higher — close to double for a similar drawdown. The cost is more losing years.
When the price rises a set distance above the open during the day, it buys at that price and sells everything at the next open. It holds overnight only. It enters only on days with upward momentum and always leaves the next morning, so it does not sit through long declines. In exchange it trades almost every day and is sensitive to fees and slippage.
All five halve the order the day after a fall of more than 15% over four sessions (added October 2026).
| Strategy | Avg return | Avg MDD | Losing years |
|---|---|---|---|
| B1 | 100.0% | −19.6% | 0 |
| B2 | 73.0% | −11.1% | 0 |
| B3 | 87.8% | −13.0% | 0 |
| B4 | 47.3% | −7.8% | 0 |
| B5 | 42.8% | −6.2% | 0 |
The numbers are far above the other families, but do not take them at face value. Stops are resolved exactly with 1-minute bars from March 2010, when SOXL started trading; before that the intraday order is unknown, so a stop is assumed whenever the low touched it. Fills are also assumed exactly at the target price, so real results lose the slippage. A cost difference of 0.1–0.2% a day changes the annual return a lot.
Two strategies built for GDXU alone (a 3x gold miners ETN). They are rules the author has been testing separately, published with the same engine (on the same prices both calculations produce the same equity curve day by day). They compute as soon as you pick GDXU, and picking one on the backtest page switches the ticker to GDXU.
Unlike the other tables this one is on GDXU ($10,000 fresh each year, 0.1% fee, 2026 through September 18). GDXU listed in December 2020, so real prices start in 2021; before that the prices are synthetic, built from the 1x ETF GDX.
| Strategy | Period | Avg return | Avg MDD | Losing years |
|---|---|---|---|---|
| GD1 | Listed 2021–2026 | 57.8% | −14.3% | 0 |
| GD1 | Synthetic 2010–2020 | 2.9% | −24.0% | 5 |
| GD2 | Listed 2021–2026 | 80.6% | −13.8% | 0 |
| GD2 | Synthetic 2010–2020 | 6.6% | −26.5% | 5 |
Read both rows. The rules were built on the listed prices (December 2020 to October 2026); the synthetic stretch is an out-of-sample check done afterwards. There, run continuously, they made 1–3% a year with drawdowns around −60% — in effect they did not work. The listed-period numbers lean on a strong run in gold miners (2025 is the biggest single year), and the rules failed on other tickers (SOXL, KORU). They also trade almost daily: raising the fee to 0.2% cuts the listed-period annual return from 60% to 47% for GD1 and from 86% to 66% for GD2. GDXU is a BMO ETN and can be redeemed early.
A daily-bar reconstruction of the latest version, V4.0 (March 2026), of the method Laoer published in his book and Naver cafe. The original text is open to cafe members only, so the rules here are reconciled from two published summaries. If you plan to follow it, check the author's own rules.
The table restarts $10,000 every year, 0.1% fee, 2011–2026 (2026 is a partial year).
| Ticker · strategy | Avg return | Avg MDD | Losing years |
|---|---|---|---|
| TQQQ I1 | 15.9% | −17.1% | 2 |
| TQQQ I2 | 19.8% | −21.2% | 1 |
| TQQQ buy and hold | 57.0% | −39.6% | 3 |
| SOXL I1 | 25.2% | −24.6% | 3 |
| SOXL I2 | 29.5% | −31.1% | 2 |
| SOXL buy and hold | 67.1% | −56.4% | 6 |
Clearly shallower drawdowns and fewer losing years than holding — and far less return. About two thirds of the money sits in cash. It beat holding the same average exposure on random days by 5–12 points a year, but trailed simply buying the same amount every day of the year (about 24% on TQQQ, 31% on SOXL). Run continuously over synthetic pre-listing prices through the dot-com crash and 2008 (1999–2010), the account fell 12–95%. 2011–2026 was an unusually kind period for 3x ETFs. Moving the sell target and star numbers by ±25% changed little (−3 to +3 points of average return). The full check is in Infinite buying V4.0 backtest.
A high backtest number does not make a good strategy, because that number may be the result of shaving parameters to fit the past. To tell the difference, the backtest page has a 🗺️ sensitivity map. It sweeps two parameters across a grid and colours every cell. Reading it is simple:
A single backtest run cannot tell those two apart. Draw the map once before you settle on a strategy.
→ Run it yourself on the backtest page
→ Longer articles in the guides
About · Privacy · Terms · Disclaimer · Contact
Last updated: 2026-10-18