Meh-mentum

In science. Proper science. Scientists are told to report the investigations that lead nowhere.

Why is this important?

One. So people don't unknowingly waste time down the same dead and can learn from the experience of others.

Two. When you count the dead ends, sometimes statistical tests change. For example, if I am testing the efficacy of a new drug. I test the drug on a thousand patients and report a p-value of 5%.

That means I have a one in twenty chance of reporting a fluke.

However, if at first I tested another slightly different drug in the same way but with a higher p-value, and include that data in the current report, the original p-value should be 10%.

Implicitly, I am reporting the best p-value from the two tests, the test becomes the probability that the first or second test is a fluke.

In any case, almost every scientist is guilty of not reporting all their investigations and due to their vanity the world is a poorer place.

Here one such small investigation I made into dual momentum.

On a monthly basis, the strategy shifts 100% of your portfolio into the best performing asset ('relative momentum') over the last year.

Once the best performing asset has been selected, it's compared to the return on a short term cash-like asset.

If the asset has been performing better than 'cash', good, if not move your portfolio 100% into cash ('absolute momentum').

It doesn't take a genius to realise that by treating cash as an asset, the rule becomes: 'select the top performing asset'.

Code for the Lazy Backtesting IDE is here.

Testing with Real Estate, Gold, Bonds, International and US equity over the past 10 years the Sharpe is 0.33. About the same as an equally weighted portfolio benchmark.

The original paper quotes a 0.73 Sharpe over 40 years, which is the equivalent of 2.3 over 10 years.

On the other hand when I backtest with just the S&P500 over 55 years the Sharpe is 0.54, which is the standardised equivalent of 1.8 over 10 years.

The upshot is that Dual Momentum hasn't performed well for the last decade, but has over 30 years beforehand.

Would you persist with a strategy for over a decade in the search for good returns?

Not likely, right?

Performance consistency is super important.

I will add some new features to that end to the Lazy Backtesting IDE soon.