In most treatments of stock selection, thematic investing (in the selection of individual securities) is often contrasted with factor investing. The term “quantitative” is often applied to “factors” identified by mathematical modeling (usually via regression modeling). It is almost as if one camp is housed in the English department, and the other in the science department. My thesis in this little paper is that the two ideas are not discrete, but rather different aspects of a continuum. In the language of statistics, there is a higher order factor that consists of a basket of “those things which cause stock prices to move.” From that perspective, it may be safe to say that themes are merely one more factor among many.
There is nothing wrong per se with identifying lower order factors and building investment models based on those juicy tidbits. Modern portfolio theorists have long advocated the use of multifactor models to capture the gains offered by each factor. If we examine the factors that are believed to have persistence and validity in the scholarly literature on factor investing, we find a few commonalities. The first is the presence of a risk premium. While there is plenty of room for disagreement, there is a plurality opinion that all persistent factors are justified by a risk premium (or perhaps the perception of a risk premium). I argue that the difference between quantitative factors and qualitative themes is merely a delimitation caused by measurement issues, not theoretically relevant ones.
Themes are by nature abstractions, and as an abstract construct, they defy easy measurement. This does not take away from the fact that hard to measure concepts can have a huge impact on social variables (I consider economics a social science). Every investor has been warned about the dangers of making investment decisions based on the powerful emotion of fear; we do stupid things when we are scared. We cannot conclude that fear is an unimportant variable because it is not easy to measure. Researchers have gone through the trouble to devise fear indices to tap into this latent construct. Investment advisors are charged with a similarly difficult task when they try to determine the “risk tolerance” of a client.
Technologists have yet to develop a working risk-o-meter, so different firms take a stab at the problem in different ways. Some are quite ingenious and seem to work well, while others are plain silly, such as asking the client to assess their risk aversiveness. This brings up the issue of measurement quality; when we attempt to measure hard to measure things, we often do a poor job of it. Social researchers tend to refer to these issues in terms of reliability and validity. Validity concerns the issue of measuring what you think you are measuring, and reliability refers to the precision of the measurement. I precise measure means a repeatable measure and a better likelihood of coming up with investable results.
Thematic Investing versus Market Sentiment
There are several measures of investor sentiment available to investors. These can be based on analysis of positive versus negative social media posts, price movements, and a host of other methods. I believe it useful to differentiate a theme from sentiment. Measures of sentiment, at their foundation, are measures of the sum total of short-term investor opinions given everything that they know about the market. My sentiment concerning Amazon stock may be heavily colored by my overall bullish take on the market in general. I may buy Amazon based on this bullish sentiment but may not hold it if markets start to sell off in general. A sentiment, then, is a short-term proxy for the collective guess as to how a particular security will perform in the short term.
A theme, on the other hand, depends on a long-term belief about a particular sector or company. Themes can be applied to sectors and subsectors, but this can often be misleading. We all know that there is a very powerful “technology” them on Wall Street, and the overarching belief is that technology will continue to grow exponentially, and the profits of technology companies will grow exponentially along with it. There are several important reasons to doubt this theme, but those that have bet against it have been crushed. Themes can be applied by investors with no regard to industry standards. Tesla may be a car company, but thematically it is a technology company and its multiple reflects this fact. Amazon may technically be a consumer discretionary stock, but in the minds of investors, the technology theme holds a powerful sway.
The problem for factor investing models is that we can’t use historical market data to parse out the different themes. Academic researchers often default to the industry standard when studying sector performances and sector correlations, but these sectors are based on business fundamentals and not the perceptions of market participants. While the former may be more intellectually satisfying, the latter is what drives share prices, and that is what matters in forward-looking models based on factors.
Some Futurist Speculation on Factor Investing
Those who keep abreast of the financial news, regardless of media, see these themes emerge, and they are often given durable names. There is the “social media” theme, which is a subfactor of the Technology theme. We can add “driverless cars,” “alternate energy,” and the “humanization of pets” to the list. The list is dynamic, and I believe it to be too long to be meaningful in the context of classical factor analytic techniques. Those tried and true methods can, however, provide us with both a foundation for our analysis and a starting point for more complex analysis. There is no doubt that Google can gather up all of the financial news from around the globe and index it with blistering speed. Were “keyword” technology adapted to isolate the market news and determine repeating patterns, themes would emerge. Of course, this would be a many step process. You would have to identify a list of common terms that don’t represent themes and parse those out, and you would have to create a synonym database, and so forth.
When a theme is identified, a handful of tickers can be identified by proximity to the theme (which Google can already handle quite nicely). For example, we could start by looking at the factor structure of FANG (Facebook, Amazon, Netflix, and Google) to see if the theme is as strongly correlated as most would imagine. Once a theme has been identified and a handful of proxy stocks that have a very high correlation with the theme have been identified, we could then get a factor score and correlate every publicly traded stock with the factor. We could also use the stalwart beta to clean up the data and subtract out the influence of overall market volatility. This systematic variance (and perhaps other sources) can be identified and removed.
Once all of this was accomplished, principal component analysis and other similar techniques could be used to identify subfactors within major themes, perhaps providing the investor with a powerful edge. We may, for example, find an “alternate energy” theme that has nested within it a “battery metal” theme that can be profitably subdivided into a “cobalt” theme. As we move down the hierarchy, the more volatile our models become. This accuracy would come at a cost of diversification, so the myriad new factors generated by the themes would have to be subjected to portfolio analysis.
As the technology stands now, only a handful of companies have sufficient resources to go searching for themes by brute force data mining techniques. Others can use factor analytic techniques to identify the validity and reliability of obvious themes. The major problem thus far is a paradigm problem more than an analytical problem. If we start to view themes as hard to measure factors–but factors nonetheless–we may discover some wonderful and fascinating things about successfully trading stocks.
[amazon_link asins=’0692783652,1260026647,0743228383′ template=’ProductCarousel’ store=’thereferencepage’ marketplace=’US’ link_id=’a68aceeb-8b71-11e8-9198-8dacdf0cb4d3′]