Buying the dip

This is not a strategy in production and not one being offered. It is an idea interesting enough to build properly, and the result was interesting for unexpected reasons. It is written up here because a test that does not work teaches more than one that does, and because most explanations of this idea stop before the part where someone checks.

The idea

When a market falls hard over a few days, it tends to bounce. Not always, and not for long, but often enough that you can write rules around it. The intuition is that a sharp drop is usually more selling pressure than new information: forced liquidations, stop-losses cascading, people reacting. Once the selling exhausts itself, price snaps back toward where it was.

This is the opposite of a trend strategy. A trend system buys strength and holds. A dip system buys weakness and lets go almost immediately. Both can work, because they are harvesting different things.

How the system actually works

It runs on the Nasdaq-100 and needs three decisions: when to buy, how much, and when to leave.

Measuring the fall. The gauge is a short-lookback oscillator that reports how one-sided recent trading has been. The short lookback is the whole point: it is deliberately twitchy, so that a couple of sessions of near-continuous selling registers immediately. A slower, smoother measure would average away the event you are trying to catch.

Scaling in, not jumping in. This is the genuinely clever part. You do not know in advance whether a dip is a one-day wobble or the first day of a 30% drawdown, so the system refuses to guess. It buys in three legs, one per day, at progressively more extreme readings on that gauge. Each leg is a third of the capital. A shallow dip that recovers quickly only ever gets a third of the money. A deep one gets all of it, at three progressively better prices. The ladder converts a prediction problem into a sizing problem.

Leaving on the first sign of life.The exit is a single rule: the first close above the previous day's high sells everything, all three legs at once. No target, no trailing stop. The average holding period is about five days. This feels far too quick, and it is the rule people most often break. The bounce is what you came for; anything past it is a different trade with a different edge, and holding for it quietly turns a mean-reversion system into a trend system that is bad at its job. A ten-day time stop covers the case where the bounce never arrives.

The interesting problem: when do you switch it off?

Buying dips in a bull market is a fundamentally different activity from buying dips in a crash. In 2008, every dip was followed by a bigger one. So the standard fix is a regime filter: only buy when the index is above its 200-day moving average, and stand aside otherwise.

It works, and it creates a strange problem. With that filter on, the test placed zero trades in 2001, zero in 2002, and 21 in 2008. The strategy switches itself off in precisely the years everyone imagines dip-buying is for. Twenty-seven years of history turned out to be, in practice, a long sequence of bull-market dips. The filter had not made the strategy robust across regimes. It had removed the regimes.

Turning the filter off doubles the drawdown and roughly doubles the return. Which of those two versions is “the” strategy is not a question the backtest can answer for you, and noticing that was worth more than either number.

What happened

It wins 78% of the time. That sounds excellent and means very little: the wins are small and the losses are larger, which is the normal shape of a mean-reversion system and the reason a high win rate is such a poor way to judge one.

Over the full period it made money, and it made less money than simply holding the index would have. Its risk-adjusted return was a fraction below buy-and-hold. Its drawdown was about a third as deep, because it sits in cash roughly four days out of five. And its statistical confidence was, in the most useful sense, nil: over twenty-seven years the data cannot reliably distinguish its risk-adjusted return from zero. That last point is why it is on this page instead of on the strategies page.

The part that actually mattered

The useful finding had nothing to do with the returns. This system's results are almost perfectly uncorrelated with the index it trades: it makes its money on different days, for different reasons. Blended with an index position, the combined portfolio's risk-adjusted return improved meaningfully, even though the dip system on its own was the weaker of the two components.

That is a genuinely counterintuitive result, and the reason the work is kept. A strategy that is mediocre alone can still be worth holding if it is mediocre at different times than everything else you own. It is also the reason it is not published as a strategy: tested as an overlay on the book that actually runs, the improvement was too small to distinguish from noise. Uncorrelated is necessary. It is not sufficient.

Why this is here and not there

A number goes on the strategies page when it can be defended against someone actively trying to poke holes in it. This one cannot, so it lives here instead: an idea worth understanding, which is what it is.

Everything above is a backtest of a rules-based system on historical data. No capital is committed to it. Nothing here is investment advice. See the full disclosures.