Why Omega and k-means clustering?

Research
1

Have you ever wondered why P123 likes Omega and k-means clustering?

Let me show you a quote explaining this:

[quote]
I myself have found most econometric tools seriously flawed, and am continually astonished by the elementary mistakes academics have made. William L. Sharpe, for example, confounded beta, which measures the slope of a linear regression, with a simple multiplier, and came to the completely nonsensical conclusion that higher beta would imply higher alpha. (I actually published a mathematical proof that lower beta results in higher alpha.) Academics place a tremendous value on standard deviation, which I think is an almost completely useless piece of data.

Yuval
[/quote] Emphasis is mine.

Omega and k-means clustering do not use standard deviation.

We will never see a method that uses a a standard deviation, including a p-value (which uses a standard deviation) or any standard cross-validation technique (that uses standard deviation) make it through the forum into production.

You will never see an information ratio (that uses standard deviation). Probably nothing new that uses a Sharpe Ratio (that uses standard deviation, Plus William L Sharpe was just not good at what he did).

Basically you will never see any mainstream technique make it through the forum into production at P123.

No techniques used by AQR or Patrick O’Shaughnessy.

Are you good with that?

It might be okay if we had someone with a statistics degree to find the suitable nonparametric statistics replacements. We do not,

It might be okay if we had someone with a financial degree and we weren’t just using the pretty buckets and tens-of thousands of Excel spreadsheets to guide us. We do not. Marc is just an occasional visitor to the site.

We have none of that. We are a cult: “The Alpha and the Omega.”

WHATEVER YOUR BELIEF I THINK WE HAVE RECEIVED THE FINAL WORD FROM P123.

-Jim

2

Omega? There have been a couple of times when I looked at Omega and so far have been unable to grasp it. I’ll form an opinion if/when I ever come to understand what it is. As to dk-means clustering, that’s a step behind Omega for me because I’ve never even seen a definition or example. But ultimately, same answer: I’ll form an opinion if/when I ever figure it out.

As to the Yuval quote, I am fully familiar with the limitations of academic work but don’t share the the extent of his beliefs, and we’ve had conversations off line.

Getting to what I think is the point of this post, we aren’t a full-fledged high-end quant shop so no matter what we do or don’t do, the most advanced quants here are likely to still have the sense that there are things missing, just as Primus (see the data drill-down thread) is disappointed that we can’t max out in the direction of data detail. I hope you’ll understand that like any business (including the ones you find and invest in through your models), we can’t be all things to all possible users and must allocate or resources as best we can. One may disagree with some of our decisions along these lines, and we may wind up getting some wrong. But one thing you can be absolutely sure of is that we hear you and try as hard as we can to act in good faith.

One noticeable omission, information ratio, is, at least in my opinion, something we do need to add and when I finish this post, I’ll go into our internal project management platform and add it.

3

Marc,

IMHO, you would be better served looking into mainstream methods.

Personally, I am okay if the mainstream methods are developed slowly or not at all.

I just do not like being lectured about the wonders of fringe methods WHILE mainstream methods are being rejected.

Omega is not worth looking into IMHO.

P123 doing k-means clustering would be RIDICULOUS. Please, do not associate me WITH that idea. I am AGAINST it.

But we have to hear about Theil-Sen, k-means clustering, Omega while cross-validation, p-values and standard deviation are rejected out of hand.

If we use add any methods look at what Patrick O’Shaughnessy (for example) is doing. NOT SOMETHING THAT LITERALLY NO ONE IS USING. Basically, no one uses Omega or k-means clustering.

-Jim

4

Just because I like something doesn’t mean P123 likes it. Never once have I ever suggested that P123 would offer omega or k-means clustering or Theil-Sen slopes. In fact, it has never occurred to me that P123 would ever offer these measures because they’re not mainstream.