Buying what everyone just sold

The first is RSI, a score from zero to a hundred of how hard a price has been pushed recently. Below thirty-five means the pushing has been downward and fast. The second is the Bollinger band, a line drawn a usual distance below the twenty-day average price. A close under it means the price is further below its own recent average than it normally strays. Both at once, and the rubber band is stretched.

It sells on either of two recoveries: RSI back above sixty, or the price back up to its twenty-day average. Neither waits for the old high. The trade is for the snap back, not for the whole journey. A stop a few percent below the entry handles the case where the fall was not a dip at all, and each position is sized so that hitting it costs a fixed small fraction of the account.

This is the busiest of the share strategies by a wide margin. Over five years and ten companies it made well over two hundred trades, roughly one a week, and about seven in ten of them made money. The average win was small, a fraction of a percent of the account, and the average loss was nearly twice the average win, which is the shape mean reversion always has: many small snaps back, and the occasional dip that turned out to be the start of something.

The margin between money won and money lost is thinner here than on the trend strategies, and that is worth sitting with. A rule that wins seven times in ten and still only comes out modestly ahead is a rule whose losses matter. The stop is what keeps them from mattering more.

Seven trades in ten win. The three that lose are each nearly twice the size of the seven that win. That is the whole business model, and it is a thin one.

The conservative version adds a filter: it will only buy a company that is still above its two-hundred-day average, so it buys dips in rising companies and refuses to catch a falling one. That is a sensible rule, and it is the one most people would add if asked. It also made the strategy so selective that over five years it found too few trades to judge and finished almost exactly where it started. The quality bar needs a record, and there was not one.

We left it that way rather than loosening the entry to force a number out of it. Loosening a rule until the test looks good is the oldest way to produce a strategy that only ever worked in the test. If the conservative version is listed one day, it will be because the design changed, not because the threshold did.

The strategy holds shares overnight, because recoveries take days and a rule that sold everything at the close would sell every trade before it worked. Overnight is where a company can announce something and open ten percent lower, straight through the stop. The test charges that when it happens in the history, but five years of ten large companies contains only a few such mornings, and the next five years may contain more. It is the one risk here that a wider stop does nothing about.