Author: The CFA UK Investment Studies Working Group
The CFA UK Investment Studies Working Group had a chance to catch up with Kai Wu from Sparkline Capital based in New York. Wu has an innovative approach to value investing which we explore further.
Value investing has been around for some time and has built a reputation of a distinguished pedigree. Value investing, as developed by Graham and then later by Buffett and Fama, has a strong fundamental reasoning behind its reputation and it might have become controversial to think otherwise.
Wu argues that classic value investing has some implicit bets that investors should be aware of it. In his piece Value Investing is Short Tech Disruption - August 2020, Wu isolates the exposure to technological development, using Machine Learning (ML) and he concludes that this fully explains the multi-year underperformance of the value factor.
As technology continues to reshape industries, Wu believes value-based investors need a new framework that accommodates the rising role of the digital economy. We asked him about his research findings and implications for investors.
Q: Kai, in your blog “Value Investing Is Short Tech Disruption” you make the interesting observation that the multi-year underperformance of the value factor is due to an implicit bet against companies that use disruptive technology. It is common wisdom that the value factor tends to short companies from the tech sector, but your point is about disruptive technology. Can you explain to us the difference and how you’ve reached that conclusion?
A: It is important for investors to recognize the limitations of industry classifications such as GICS. These taxonomies are rigid and reductive. Only 20% of the FAANG are actually considered IT companies. In reality, companies can operate across multiple industries in a much more fluid way. Disruption is a great example. While many tech companies are disruptive, there are also tech companies employing legacy business models. Conversely, there are many companies that are disrupting old-school industries such as industrials and financials.
I used machine learning to classify companies as “disruptive” or “non-disruptive” based on the themes extracted from a variety of text documents. I found that value investors have maintained a significant anti-disruption bias the past decade. This has been a very costly bet given the significant transformation we have witnessed over this period. In fact, when we neutralize this bias, we find that the entire value v. growth drawdown disappears.
Q: Trying to incorporate intangibles into traditional valuation metrics is not a recent field of study, but it is becoming a trendy topic. Why do you think this is important today? And does adjusting the intangibles component is enough to understand the intrinsic value of a company?
A: The intangible economy has been on the ascent for multiple decades. While intangible capital was a rounding error in 1980, I estimate it is now almost half of the capital stock of U.S. companies. Thus, the trendiness of the topic is completely justified. The problem with accounting is that is largely ignores the rise of intangibles. This creates issues for investment strategies that rely on financial accounting metrics such as price to book ratios.
Starting with Baruch Lev, many researchers have suggested the capitalization of intangibles such as R&D. While I believe this is a starting point, the problem with intangibles is they are inherently uncertain. Thus, while accounting reform is helpful, we need to go far beyond this if we hope to truly understand the intangible economy.
Q: Recent academic research, like Rizova and Saito (2020)1 suggest that there is no compelling evidence that investors should include estimates of internally developed intangibles in company fundamentals such as book equity. Most of these estimates are very noisy. Do you think this contradicts your conclusions? What are they missing? Or how should we balance the challenges of capturing intangibles with understanding the value of a company?
A: As in any healthy debate, there will be compelling arguments on both sides. We can balance their findings with those of practitioners (including myself), who find that including intangibles slightly improves the performance of price to book. I find the argument for consistency to be compelling. In other words, we should treat intangibles and tangibles, externally- and internally-generated investment the same. However, the effect of these seemingly sensible adjustments on investment performance is limited. I believe this is due to the inherent uncertainty in the output of intangible investment. Therefore, valuation of intangible assets based on historic accounting cost is only a starting point.
Q: From your comments, it seems that investors need to go beyond accounting metrics to properly estimate the intrinsic value of a company. In your blog you suggest that Machine Learnings techniques and Natural Language Processing can be helpful. Can you explain to us why you think they are important going forward and how you have used them so far?
A: I think machine learning - and natural language processing in particular - will eventually become commonplace in the investment industry. I am less interested in how ML can be used for technical analysis. To me, the promise of ML is that it allows us to incorporate unstructured data into the investment process in a systematic way. Unstructured data (text, images, etc.) comprise 80% of data and this is growing. Ultimately, this can be used to automate many of the tasks that fundamental analysts do on a daily basis. Even more exciting is its potential to extract information that would be impossible for a human. In addition to using ML to classify disruptive companies, we can use it to gauge CEO sentiment on earnings calls, perform automated topic modeling on news, and improve our risk models by capturing a deeper understanding of the business landscape. The use cases will increase over time as data collection, algorithmic design, and computing power continue to increase.
Q: What is the impact to individual investors in a world that depends more on these techniques for successful investing? How can you or others who manage money get the Machine Learning component right when there are some companies with large budgets and teams and they can’t still crack it?
A: One of the great things about technology is that the barriers to entry are constantly falling. At one point in time, only MIT or Bell Labs scientists had access to mainframe computers. Today, it is easy to get access to vast computing resources through AWS, Azure, or Google Colab. Cutting-edge ML libraries such as PyTorch and TensorFlow are open source and available on Github. Excellent ML training resources exist online, notably Fast.AI. The binding resource is not money (above a threshold), but creativity and insight.
Q: For years, investors and academics have valued tech-related companies and many others. And they have done so without data science techniques like ML. There are ways to value companies like Amazon or Tesla with a dividend discount framework. Hence, do you think the critique you do in your blog is more about a mechanical bias in the value factor rather than the actual value investing philosophy?
A: For the record, I do not think value investing as a concept is broken. Buying things below intrinsic value is still a good idea, obviously. The problem is with the definition of intrinsic value. Think about the Fama-French value factor based on price to book. Book value is simply not a good measure of intrinsic value (perhaps it never was?). However, I think if value investors can adapt their metrics to include some of the things I’ve mentioned, such as intangibles, cries of the “death of value” will start to go away.
Q: In your blog you use the example of Buffett as an investor that doesn’t follow simple mechanical rules, but one who applies value investing principles to the current investment and economic landscape. Can you explain further what you mean by this? Is Buffet an investor who would buy companies at high multiples like P/B?
A: I like the career of Buffett because it illustrates the value of adaptability. Too many investors develop a rigid framework and then struggle when the world changes. Buffett has been in the game for decades and has found a way to change with the times. He started off following Ben Graham-style investment principles, which he described as “cigar butt” investing. As the industrial economy gave way to the great American brands, he further evolved to a “quality” style with the help of his partner Charlie Munger. Most recently, Buffett has described the economy as “asset-light” and put 20% of his portfolio into Apple due to the strength of its “ecosystem”. Another lesson is that Buffett understands intangibles such as brands and network effects. As he would say, these are the deepest moats.
Q: What skills do you think CFA UK members should develop in the next few years in order to keep up with the pace of innovation and excel as financial analysts? What would your advice be for active investors, based on your findings?
A: I think the key skill is to have an open mind. Skills can be learned, but the key is not to get stuck in ones ways. But to be more concrete, I would recommend learning the Python-based data science stack (Numpy and Pandas). Even these simple data science techniques will give you a big leg up over other investors. I would also spend time understanding the business models of tech companies. Even if you are not a tech analyst, technological adoption is spreading across the economy broadly and will change the dynamics of many industries. It holds the key to disruption.
Q: And finally, my last question, what research topics are you currently focusing on?
A: I am working on a paper on network effects. Technology and globalization are making the world smaller, thus exponentially increasing the potential value of networks. Companies like Google and Facebook have taken advantage of this tailwind to build extremely valuable businesses. I am thinking a lot about the role of narratives in investing. Finally, I have been very interested in the Chinese economic development story for many years. In particular, the rise of their own tech ecosystem and how it plays into geopolitics
Kai Wu is the founder and chief investment officer of Sparkline Capital.
Sparkline Capital are an investment management firm applying state-of-the-art machine learning and computing to uncover alpha in large, unstructured data sets.
Prior to Sparkline, Kai co-founded and co-managed Kaleidoscope Capital, a quantitative hedge fund in Boston. With one other partner, he grew Kaleidoscope to $350 million in assets from institutional investors. Kai jointly managed all aspects of the company, including technology, investments, operations, trading, investor relations, and recruiting.
Previously, Kai worked at GMO, where he was a member of Jeremy Grantham’s $40 billion asset allocation team. He also worked closely with the firm's equity and macro investment teams in Boston, San Francisco, London, and Sydney.
Kai graduated from Harvard College Magna Cum Laude and Phi Beta Kappa.