Friday, October 16, 2009

Why I am not a bottom-up equity quant

I spent close to two decades of my career building bottom-up equity quantitative models to pick stocks, in Canadian, US and international markets. I was asked why I don’t do that anymore.

My flippant answer was:“Been there, done that.”

My longer answer was that the competitive advantage of doing bottom-up quantitative analysis is being eroded to such an extent that alphas are rapidly diminishing.

Let me explain. Back in the 1970s and 1980s, the task of performing equity quantitative analysis required a large commitment by an investment organization. Sure, there were databases around, but the task of integrating them was a non-trivial task that required investment in staff and infrastructure.

Here are some sample issues. How do you marry an earnings estimate database (e.g. IBES) with a fundamental database (e.g. Compustat) when:

  • The series have different periodicities (annual & quarterly for the fundamental and daily/weekly/monthly for earnings estimates)?
  • The identifier for earnings estimates is for the security (stock specific) but the fundamental database is identified by company (as multiple share classes are not uncommon for non-US companies)?

Add to that the issues of adding data for new listings, deletions, name changes, etc. The investment organization quickly finds itself not in the investment business, but the database maintenance business.

Falling barriers to entry
Fast forward a couple of decades, the apperance of system integrators like Factset Research Systems have revolutionized the business and dramatically lowered the barriers to entry to bottom-up equity quantitative analysis. Today, you can build an equity quantitative research capability by subscribing to these services.

Opera singers don't belt, quants control for factor risk
Moreover, a generation of quants has been conditioned by the likes of Barra to decompose risk as industry plus a Fama-French like common factors such as Market Capitalization and Style (Value/Growth).

The implications of this analysis framework is that just as opera singers are genetically imprinted not to belt when they sing, bottom-up equity analysts are conditioned to believe that you shouldn’t try to forecast the returns to these risk factors. Instead, the appropriate way to forecast alpha is to forecast alpha based on residual risk, or stock pick after controlling for, at the very least, industry and sector risk.

The combination of lower entry barriers and groupthink has led equity quants into a crowded trade. They all uses some form of multi-factor stock selection model, but the data comes from the same databases. The factors all appear to be uncorrelated but we saw what happened in August 2007.

The low lying fruit is gone
Even when you succeed, it’s a really tough business.

I recently attended a seminar put on by a risk model vendor and a respected equity quant manager. The equity manager put up an analysis showing various ways of integrating their forecast alphas with the risk models that they use. The most optimal technique for a long only portfolio, with a 2-2.5% forecast tracking error, resulted in an annual alpha of about 1.5% a year.

1.5% sounds pretty good.

However, you have to consider that this is a forecast alpha from a model portfolio with no turnover costs. Once you throw in trading costs, the shortfall between the turnover of the forecast alpha and the actual portfolio, which could vary greatly, and even the fact that good managers with tight investment processes experience portfolio dispersion (difference in returns between accounts with similar mandates) of 2% or more, 1.5% doesn’t sound that good.

As I said before, it’s getting to be a really tough business.

So far my solution has been to do something that is truly queasy and nauseating to many equity quants. I have been using top-down investing and factor rotational approaches to quantitative investing. The approach disturbs quants because it's less disciplined, less risk controlled (according to the way they are trained) and appears to be so, well, empircally oriented.

My approach is not the only answer but bottom-up equity quants need to find new sources of alpha.


Henry Bee said...

Utterly and completely agree.

jhunt said...

Great post Cam. Especially love the emphasis on 'empirical' at the end. As a graduating finance student, it's saddening (but not surprising) to see that the 'tricks of the trade' we learned from our upper level classes were taught to all of us. A crowded trade indeed.

jhunt said...

Great post Cam. Especially loved your emphasis on 'empirical.' Such a dirty word to some people who would rather lounge in the comfort of numbers.

It's sad but unsurprising that all those tips I've learned in the past four years of school are for naught. Perhaps finance as a whole is a crowded trade?

Unknown said...

Really good piece and i mostly agree with you. Yes, alpha is very hard to come by nowadays given the barriers to entry have been significantly reduced because of the commcially available, turn-key solutions that you mentioned. Additionally, in a world of a few risk model and data provider dominating the field, crowded trades also cause the correlation to rise and even harder to get orthogonal alpha for managers. However, the bottom-up quants are still kicking around seeking new sources of alpha as you said at the end of your post. And the managers who are fast and innovative will get there first to harvest them. Then the game will change again. Quant investment is a fierce competition on generating new ideas, on technology, on speed of implementing new strategies. It is multi-dimentional. It has always been that way. The difference is the cycle just got shorter and more volatile. But make no mistake, there are still plenty of alpha to be captured using bottom-up quant. The question is how and who gets it first every time. By the way, transaction cost can also be a new source of alpha, rather than just cost which can be incorporated and captured in a bottom up quant investment process, longer dated research is definitely warranted on this topic. So apparently, i am still a bottom-up equity quant. :)

Patrick said...

I think the big alpha opportunity for the next several years will be exploiting model risk.

keithpiccirillo said...

What type of benchmarks have priority for most quants when they rank stocks?
Is it valuation, sentiment, or prospects for earnings?
In this environment does one favour value or growth, large cap or small?
One would think most value stocks and the risk trade have sufficiently been picked over and large caps should lead going forward.
Rhetorical questions but when I see hedge funds like Tudor zipping in and out of tiny African energy and water plays it is mind boggling for the average fund manager to do well.

Cam Hui, CFA said...

Keith -

A standard equity quant multi-factor model would typically have the following kinds of factors in it:

- value
- growth
- growth at a reasonable price
- momentum, either fundamental, e.g. estimate revision or earning surprise, or price momentum
- quality (earnings or balance sheet)

I know some of the guys at Tudor and they tend to be more top-down oriented and not the disciplined quants in the classic sense. Their rotation in and out of illiquid names are probably a function of some other form of investment insight.

Unknown said...

All we have to do is look to the July-August period as the defining moment of implosion and anyone with their ear to the ground would have heard the R.I.P. to Factor Backtesting...but that is NOT a RIP to bottom-up equity quant! The problem is not the GOAL it is the's tainted from the get-go. What some predicted would happen actually happened...and you said it correct; everyone doing the same factor back-testing with the same historical data and we drive the Alpha out of the Methodology. The best signal that multi-factoring back-testing (MFBT) was DOA was the creation of the so-called "Alpha Platform"..notably in my mind CLARIFI ...

So, let's separate MFBT and bottom-up quant techniques...what's the limitation to MFBT? First, it's not dynamic but rather totally static..'what happened in the past will repeat itself"...when the fundamental change or the market (i.e prices) change the relationships no longer exist. Second, it is a biased starts with viewpoint, perspective or belief in certain's not quant it's tests these relationships over time, validates the bias and "ta-Da" we have have an 'edge"...WRONG! Over time the rest will catch up drive the Alpha that point you should be short the MFBT signals !

What is needed is a Dynamic methodology that adapts when the fundamental data changes or the market changes. One that is data agnostic and holds no preconceived bias as to what the drivers of value are...and one that seeks the drivers of value at a more granular, and at a relative basis to skim off the obvious and to identify the under AND over valued equities.