Value Investing: Robots versus People
Who is better at value investing: robots or people? How have robots – the quantitatively-driven passive funds that hold, for example, low price-to-book stocks – fared against actively managed value mutual funds?
A provocative paper forthcoming in the Financial Analysts Journal, “Facts About Formulaic Value Investing,” by U-Wen Kok, Jason Ribando, and Richard Sloan, argues that robots are poor value investors.
You need people – analysts studying companies – to pursue value strategies successfully, argue the authors. The reason is that robots have no intuition for the data and can only take the data literally, while, in value investing, the key to success is in knowing when certain types of data, such as earnings and book value, can be trusted and when they are exaggerated or manipulated.
But is this true? The article is erudite and well-crafted and refers to all the right people. For example:
More recently, Asness, Frazzini, Israel, and Moskowitz (2015) reached a similar conclusion about the performance of standalone formulaic value strategies. Recent research has also shown that the performance of book-to-market and other formulaic value strategies in the United States has become weaker since their initial publication (see Asness et al. 2015; Fama and French 2016; McLean and Pontiff 2016)….
But active managers have had great difficulty beating their benchmarks on average over long periods of time, in practically every capital market. If active value investors are the exception, beating robotically-constructed value benchmarks with some degree of consistency, we’d like to know that.
To evaluate the claim that robots “stink” at value investing, I begin by applying a formula inspired by the great finance professor and Goldman Sachs partner Fischer Black, who believed that most financial research was data mining. Here is the formula:
- Don’t look at the data.
- Think about the question.
Value investing is about comparing a company’s price to some measure of its fundamental value, buying the cheap companies, and holding them until you’ve made a profit (if you ever do). This process can be mimicked by a value benchmark, which performs the above-described tasks robotically but which is subject to Kok et al.’s critique. So we can test the idea that robots stink at value investing by comparing the long-term performance of a value benchmark to that of an unbiased universe of active value managers.
No one is suggesting that a robot is better than all human value managers. In any investment category, there will always be a lucky or extraordinarily skillful few who beat all relevant benchmarks for long periods of time. The trick is figuring out, in advance, which managers these will be.
Is value investing a good idea?
Asking whether robots or humans are better at value investing is very different from asking whether value investing is a good idea. That question can only be answered by determining whether the conditions that caused value to beat growth historically are likely to be repeated. These conditions are usually believed to be behavioral: investors overreact to both good and bad news about companies, causing popular stocks to be overpriced and unpopular ones to be underpriced.
Let’s first briefly ask whether value investing is a good idea, then turn to the question of robots versus people.
Exhibit 1 shows the cumulative total returns on value and growth benchmarks constructed by Fama and French . Starting in July 1926 they classify, each year, all large-cap U.S. stocks into three categories: value (the 30% of stocks with the lowest price-to-book ratios), growth (the 30% of stocks with the highest price-to-book ratios) and neutral or core (the remaining stocks). The value benchmark outperformed the growth benchmark by a cumulative 7.3-to-1 ratio, or about 2% per year.
Cumulative total returns on Fama-French U.S. large-cap value and growth benchmarks, July 1926-March 2017