6 Comments

Well said. if I may ask, why is out of sample data, the data you’d be most concerned with; given that the in-sample data covers the time series of the analysis you’re researching?

Expand full comment
author

In sample data is not relevant because it is in the Ort. When it comes to making money you need a model to work out of sample I.e. with data that hasn’t been a Aula ke when you built the model. So many funds and ETFs etc are built and look good in the nacktest. But once they start run ing in the real world they fail miserably and have much weaker performance.

Expand full comment
May 19, 2022Liked by Joachim Klement

Until next Tuesday I will continue to avoid predicting anything.

Expand full comment

No silver bullet

Expand full comment

would it be valid to test a model that reduced draw downs rather than beat the market? What if I sold when the market dropped 2% below the 200ma and bought when the market rose 2% above the 200ma? Over long time horizons I would be less likely to panic sell.

Expand full comment
author

Men Faber has tested that model and yes, it works. But optimising a model to reduce drawdown is tricky. After all, in a multi-asset world, such a model will always push you 100% into cash because cash has 0 drawdown. Here is the link to Meb's paper on the 200D MA: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=962461

Expand full comment