Value investing in the age of machine learning
The advent of machine-driven investing means that computers can quickly pounce on stocks that the market seems to have mispriced. This development has raised the bar for value investors seeking to outperform the market. But it’s not the beginning of the end for human stock-pickers.
The power of computer-driven analysis was illustrated for me when I attended a recent session with Mark Grinblatt, Distinguished Professor of Finance at UCLA Anderson School of Management. He was discussing an academic paper he has written with Söhnke M. Bartram from Warwick Business School.
The paper, titled Global Market Inefficiencies, sets out to quantify the level of inefficiency in a range of global equity markets. Given that market efficiency is an enemy of outperformance, this sort of insight is helpful in thinking about where an investor might most profitably apply their capital or investment skill.
The conclusions reached are interesting, but what I found really intriguing was the approach taken by Grinblatt and Bartram in gauging inefficiency.
What they did was to predict company fair values using a regression with market capitalisation on the left hand side, and accounting data on the right hand side. They repeated this process on a monthly basis with a sample of several thousand global stocks, and a library of 21 of the most commonly-reported accounting variables.
This process generates an equation that relates accounting variables to market capitalisation. You can then feed in the accounting data for any given stock and the equation tells you what the market capitalisation for that particular stock ought to be. If the actual market capitalisation is above (or below) that number, then prima facie you have a stock that is expensive (or cheap).
There are a few things that are interesting about this:
- Firstly, there is no theoretical concept of value underpinning it. The regression equation simply reveals what the market seems to be saying about the relationship between accounting numbers and market value. If, for example, the regression indicates that the book value of assets is not important, but that tax expense (or some other variable) is, then our notion of value needs to mould to fit that.
- Secondly, the equation changes over time. Value is not a fixed concept, but instead reflects the particular accounting variables that the market appears to consider important at that point in time.
- To put it another way, the equation is not aimed at estimating underlying intrinsic value, but rather at estimating where the current market price should be.
These ideas are a little consternating to value purists, who hold that value is a fixed concept that can be uniquely defined for a given set of cashflows and discount rates. Adding to this consternation, the results presented by Grinblatt and Bartram indicate that the approach appears to work quite well, albeit less so in the more developed global markets.
The paper also highlights an interesting contrast in the way humans and machines approach investing: A human analyst might focus intently on a single company and a small set of key questions to try to gain unique insight into the possible distribution of future cashflows for that particular company. A machine, on the other hand, will simultaneously look at several thousand companies and “all” of the associated data to gain insight at an aggregated level, without any deep insight into any one particular stock.
The approach outlined in the paper is an interesting one, but it’s fair to point out that what Grinblatt and Bartram call “fair value” is actually “fair price”, and this is a very different beast to intrinsic value, notwithstanding that it calculated by reference to company fundamentals. Indeed, the fact that there is a difference between value and price is what allows value investing to work.
So, what do we take away from this? For me the key points would be:
- In skinning the investment cat, there is more than one way to think about analysing company fundamentals.
- While it is not the only path to success, investing based on intrinsic value continues to have strong intuitive appeal, and investors who are good at assessing intrinsic value can expect to succeed long term.
- However, progress and technology are raising the bar in terms of the skill needed to generate outperformance. As machine-based investing develops, obvious mispricing will more quickly get pounced on by computer-driven strategies that can scan the entire market in the time it takes me to read the Financial Review.
- As a result, investors need to evolve to maintain their edge. For example, if a good computer model can quickly and easily produce a list of companies that appear to be mispriced, a smart value investor will make sure they know what’s on that list.