My subject line is taken from an article written by the PhD Head of Investment Research at Dimensional Fund Advisors (DFA), but I cheated and left out the second half of the sentence.

Here’s their full headline: We Found 30 Timing Strategies that “Worked” and 690 that Didn’t.

I guess I’m not above clickbait, but I also don’t have enough space to get the full headline into the subject line.

The full headline is so powerful because it tells the story of market timing strategies – you can find them, but they probably don’t mean anything.

There is so much computing power compared to the amount of market data that the DFA researchers say that it’s easy to find strategies that have beaten the market.

However, if you send your data crawlers to find hundreds or thousands of strategies in the historical data set, you’re bound to find a few that work simply due to chance.

In this case, of the 730 strategies that DFA tested, only 4.1 percent of them worked.

They are quick to point out that the strategies are absurd and don’t have any of the economic intuition or finance theory that you need to have a strategy that you put money behind.

Here’s the strategy that did best in their tests, beating the market by 5.58 percent per year between 2001 to 2022.

It uses the valuation ratio to time the market premium in the developed ex US markets. At the end of each calendar year, the strategy compares the current price-to-book ratio of the market with its historical distribution over the most recent rolling 10-year period. When the price-to-book ratio exceeds the top 20th percentile of its historical distribution, the strategy gets out of the market and invests in one-month Treasury bills. When the price-to-book ratio drops below the 50th percentile of its historical distribution, the strategy switches back to the market portfolio.

Of course, that makes no sense, and I wouldn’t put a dime into it even though the back-test is great.

We do invest in strategies that we believe will outperform over time, and part of our belief comes from historical data.

However, in addition to the historical data, we’re looking for several other things besides a strong back-test.

First, we like to see a strategy work in out-of-sample data. In sample data is simply the data that was used to create a strategy. It might be limited to a certain time period or geography. A strategy that only ‘worked’ in Brazil in the 1980s won’t work for us. We want to see it work in most countries over many time periods.