Corey Hoffstein @choffstein CIO at Newfound Research (@thinknewfound). ⚡️ Risk cannot be destroyed, only transformed. ⚙️ Quant / Systematic TAA. 📚 Jan. 26, 2019 3 min read

📌 1/x Lot's of arm-chair blogger feedback on my GEM commentary ( )

It's like y'all are new here and think I've never researched this stuff before.

So I'm taking a page from @ROIChristie and just going to assemble a tweet storm here to point people to.

2/x First of all, my article was not an indictment of GEM.

It was supposed to highlight the difference between *style* and *specification* risk.

3/x My firm was founded in August 2008 to help manage trend equity strategies.

For those keeping track, that's 4 years before @GaryAntonacci wrote his paper about GEM.

So this isn't an anti-trend thing; I'm all in there. We're talking implementation details here.

4/x "Corey doesn't seem to recognize that GEM is based on 800 years of data!"

First of all, if you think we have 800 years of good, clean financial data, see yourself out.

Beyond that, I'm plenty aware of the evidence behind momentum and trend. 

5/x "Corey only looked at 10 years of data!"

That's sort of the point. 10 years is more than enough to get a professional manager fired or hired. And most retail investors quit well before that.

But, you know, I've looked at other periods as well: 

6/x "If you want actual diversification, use a different type of model!"

Great strawman argument. The point here is all about consistency between expectations and results within a style.

And intra-style differences can more extreme than between styles: 

7/x "The diversification benefits are marginal."

Depends how you measure diversification benefits.

By reduction in volatility? Sure. But even moderate diversification benefits can be a force multiplier: 

8/x "Simple is more robust!"


9/x "The combined model will trade more!"

Yes (though turnover will likely be about equal.)

That's sort of the whole point of the argument: small, frequent adjustments are less likely to cause extreme performance jumps than big, sudden ones.

10/x "The combined model will trade more!" (continued)

If the implication is about trading costs, we should acknowledge that most of the ETFs used here are zero commission and hyper liquid with extremely narrow bid/ask spreads.

11/x "It all evens out in the long run."

No. It absolutely does not if we are talking about the dispersion in terminal wealth. 

12/x "12-months is best over the long run!"

No. 12-months has proven most effective for a *particular measure* of trend following over *one particular market* in *one particular realized history*.

In other markets or shorter time horizons, this is not true.

13/x "12-month is best over the long run!" (cont.)

The Sharpe ratio of 6-, 9-, and 12-month GEM are statistically indistinguishable. The fact that 12-month had a higher realized return is indistinguishable from noise and luck compared to other models.

14/x "12-month is best over the long run!" (cont.)

It's worth further pointing out that 12M implemented at the end-of-month is just one variation of the 12M. And implementing at different points in the month have starkly different results (timing luck).

15/x "Combining multiple models together is impractical and difficult!"

If you can't figure out how to average a few numbers together, you have bigger problems here.

16/x "Investors just need to hold for the long run!"

I've never met an investor who has a 100-year horizon. Most have a 20-30 year horizon and *expect* 12M GEM to protect them in the next large drawdown.

Depending on how that drawdown is realized, 12M can work or not.

17/x "Investors just need to hold for the long run!" (cont)

The risk is *your realized path*.

See Australia in the 1980s. Or Germany in the 1940s. Or FTSE (UK) in the 1720s (ha!) or the early 1900s. See US returns in the late 1930s.

18/x Diversification within a specification is about trying to minimize the difference in expected results and realized results.

Terminal wealth dispersion measures this well and there has been plenty written on that. 

You can follow @choffstein.


Tip: mention @threader_app on a Twitter thread with the keyword “compile” to get a link to it.

Enjoy Threader? Sign up.