The One That Beat Buy-and-Hold

The SOXL post ended on a claim I could not prove: that a dip accumulator earns its keep when the price oscillates. LABU is where that got tested properly, and the answer is yes – narrowly, slowly, and not for the reason I expected.

The bot finished ahead of buy-and-hold. It was behind buy-and-hold for ten of the fourteen weeks, by as much as $161 at the worst point. Everything interesting about this bot is in that sentence.

Why LABU, and why a lower trigger

SOXL runs a −7% buy trigger because SOXL is the most volatile name in the fleet – roughly an 8.7% average daily move. LABU averages closer to 5%, so the same threshold would almost never fire. The LABU bot is tuned to −4%, which is the same idea expressed in the asset’s own units: buy the days that are genuinely bad for this ticker, not the days that would be bad for a different one.

Everything else is inherited from the SOXL bot:

• Buy 50% of available cash on any day down 4% or more against the previous close, capped at two buys a day.
• Sell in bands above average cost – 20% at +5-15%, 30% at +15-25%, 45% at +25-40%, 50% beyond +40%.
• Ratchet the next sell to +5% above the last sell price, not above cost, so a long climb keeps triggering.
• No-loss gate – never sell below average cost, whatever the signal says.
• Red-day gate – never ratchet-sell while the stock is down on the day.

The last two do more work than they look like they do. Both show up in the table below.

Adding cash on the way down

The other difference from SOXL: I kept feeding this one. It opened with $200, took another $300 after the first real drawdown, and another $500 when LABU gave back a third of its run. That is a decision I made, not the bot – but it is the decision the bot’s own state was telling me to make. A dip accumulator with no cash is just a holder, and every top-up went in when the ladder had spent its dry powder.

Three cash injections, five dip buys, six ratchet sells. Where does the ladder actually gain on holding?

Illustrative: a fourteen-week path with the real bot's rules and cash injections. Not a trade log.

$1,000 in, $1,122.75 out against $1,103.59 for putting the same three cheques straight into shares. The bot banked $153.73 in realized gains along the way and still holds 3.884 shares.

When the bot was winning, and when it was not

Dollar totals hide the story because capital went in three times. The gap between the two lines is what matters: the bot’s value minus what those same cheques would have been worth sitting in shares.

If the bot won, when did it win?

Bot advantage

Illustrative, from the same path as the table.

Every sell above cost cuts exposure, so a rising LABU pulls away from the bot by construction -- the gap bottoms at -$161 exactly at the top. The bot only crosses into the lead at week twelve, on the way back down.

What worked

Selling into strength, mechanically. Six sells fired between weeks 3 and 10, banking $153.73 while the price climbed. I would not have taken any of them. The whole reason this bot exists is that I do not trim, and the bands did it without asking.

The ratchet, not the bands. Anchoring the next sell to the last sell price rather than to average cost is what let it keep selling all the way up. Anchored on cost, the +47% sell at week 10 never happens.

Cash arriving on drops. The $500 cheque landed at $290, and the bot put $529.76 of the pool to work there. The same money at week 10’s $330 would have bought a fifth fewer shares – and every one of those extra shares is still held. That timing was my decision, not the bot’s.

What didn’t

Losing the entire rally. By week 10 the ladder had sold down to 0.419 shares – a ninth of what it holds now – while LABU was at its high. That is not a bug; it is the strategy working exactly as specified, and it costs real money every time the market goes straight up.

The +5% gate idling the bot. Weeks 2, and 11 through 13, produced no sells at all: either the price was under average cost or the gain had not cleared the gate. Four of fourteen weeks, the bot’s only move was to buy or wait.

Being right about the wrong thing. The spread went positive at week 12 because LABU fell 15%, not because the bot did anything clever that week. Measure this on week 10 and the post is titled differently.

---

Next: if the buying is the part that works, the trigger is the part to fix. A bot that can only buy on a −4% day spends most of the year with its cash idle.