Thoughts on Factor Investing

The question I get asked the most during the past twelve months is “Why are factors not working?” Here are my top 12 personal thoughts on the topic—informed by 15+ years of successfully “factor investing”.  

1.       There is no such thing as factor investing.

There is only investing. 10 years ago, the term ‘factor investing’ did not exist while the underlying ideas and approaches existed for centuries. Quantitative investing, like all investing, is based on human ideas, not abstract ‘factors’ that magically generate returns. Even the king of all factors, the ‘equity market factor’ does not automatically generate a premium as there are dozens of equity markets which were either completely wiped out, or did not outperform local bonds for many decades in a row. Instead of expecting magic return premium from a ‘factor’, it seems much more prudent to think about it all as regular investing. Is your investment philosophy to buy cheaper companies with rising stock prices? How and why do you expect to have an edge against other investors in applying this popular philosophy?

 2.      Outside academia, there is really no such thing as a factor.

There are equity characteristics. Some are profitable, others are not. Just because a characteristic produces correlation,  it does not imply a ‘risk premium’. Stocks in the same industry have extra correlation but it does not imply they deserve a premium. The two characteristics that have the strongest academic backing as ‘factors’, beta and size, do not hold up empirically. The other two most popular characteristics: momentum and quality, have no risk factor theories behind them but work around the world regardless. Moreover, asset characteristics can be simple or complicated, traditional or alternative, normalized or lagged, categorical or continuous, normal or uniform, contextual or dynamic, multivariate or univariate, two-hundred-year-long or recent - all of these are personal stylistic decisions of the investor with an infinite number of permutations, innovations, and outcomes, some of which are positive while others are random. The simplified ‘factor investing’ conversation sounds a bit naïve and mid-guided compared to true quantitative investing.  

3.       The Mean, not the variance (link)

The most important risk, in modern day active investing, is the lack of a positive alpha, not the variance. I propose to stop talking about the other risks and spend 95% of the time addressing the ‘mean’ risk. ‘Factor investing’, is about strategy risk. Will the specifically chosen strategy work going forward?

 4.       Radical Innovation (link)

The only way to protect against the mean not being positive is to develop an edge through radical innovation.

 5.       Type 1 errors are a trap (link)

The focus on Type 1 Error and Data Mining is overblown and suffocates innovation. Adding a random signal into a model is not great, but definitely not as awful as popular opinion currently believes.

 6.       Type 2 errors are the problem (link)

By contrast, large Type 2 Errors, that no-one is talking about, are actually the big problem. Failing to innovate and come up at least with some signals that actually deliver positive value, is the real reason behind poor results.

 7.       Investing ought to deliver value (link)

10 years from now, no-one will be talking about “factor investing”. They will be talking about some other investment approaches that deliver value. Some of them might use valuation and momentum concepts. Others might utilize alternative data or machine learning. Others, some completely new way of fundamental investing. The rest will stick with index funds.

 8.       Where Does Alpha Come From? (link)

Personal investment style and architecture-like thinking is where alpha comes from, not textbooks, manuals, academic papers or competitors.

 9.      Attribution vs Investing (link)

Attribution models are not the same as real time investing.

10.   Then and Now (link)

15 years ago the quant-factor world was dominated by the “big few” with dozens of PhD’s that looked invincible - and they had sophisticated versions of value, momentum, quality models. Then they blew up in 2007 and 2009. After that, the industry moved into transparent, single factor, rubberstamp versions of the same models. They sold much better, but to anyone back from the early days it clearly looks like retrogression. Instead of more innovation we remained complacent. Instead of more alpha and differentiation, we found less. Expecting out-performance from these simplified products, when even their most advanced versions of the past did not deliver value, does not make sense. Some of these products that contain real differentiation (for example these guys), have a chance of winning in the future, but not simply because of the underlying factors, but rather because of their unique way of implementing ideas they contain that are aligned with their personalized investment philosophy instead of another ‘me too’ strategy.

 11.   Art vs Science (link)

Quant and all other investing is much more of an art than a science. A creative quant architect and not any given factor, is at the core of a successful quant investing strategy. I believe investing quantitatively based on company characteristics can be done successfully and there are a few notable examples out there. But the builders of such models typically are not thinking in terms of risk factors, premia investing, financial theory of compensation for risk, etc. They are thinking in terms of personal innovation style aimed at using technology to do advanced financial analysis via characteristics that lead to some overall model edge, producing some accuracy of predicting future out-performance.

12.   Opportunity (link)

The overall quant / factor model R-squares remain extremely low, ranging from 1% to 2% - which means that there is still at least 98% to 99% of variance that is unexplained. There is a vast world of ideas about what might drive future company performance. When these ideas are measured, they become characteristics. Figuring out which characteristics to study, how to define them, how to measure them, and how to combine them with other characteristics is the essence of quantitative investing. Value, momentum and quality (VMQ) are nice to have as general categories of characteristics; however, inside each of these categories are a massive number of ways they can be implemented. And, most importantly, beyond the VMQ, there is an even greater number of ways to build other kinds of models. Let’s think creatively.