Phillip Wool, PhDScroll down
Emerging markets pose challenges for many investors because their idiosyncratic features don’t often lend themselves to the standard research and investing approach. In this Q&A, Doug Gratz, CFA, Rayliant’s Director of Institutional Services, spoke with Phillip Wool, Ph.D., the firm’s Head of Investment Solutions, to understand how localization, although laborious in many ways, offers sharper insight into emerging markets and how fundamental research can effectively complement quant methods.
“Localization” is the term we’ve come up with to describe the unique process by which we try to tailor our research and development of quant models and the implementation of our investment strategies to distinct features found in the markets where we’re trading.
Much of our research is focused on emerging markets (EM). When we look at EM around the world, what we see are a group of countries sitting at different points in their financial development life cycle. Because of that, we find they have all of these weird, idiosyncratic features—things like heavy state ownership in the case of China or complicated business groups in South Korea; in India there are extremely strict limits on foreign ownership in many cases. All of these country-specific differences create opportunities to take a slightly different approach in each country to capture more of the alpha opportunity we believe is waiting for us in these markets.
You can probably think of this by contrasting with what I’d call a “one-size-fits-all” approach to quant investing. That’s where you may have an Emerging Markets low risk strategy—to take a hypothetical example—that just applies a standard low risk factor to a portfolio full of EM stocks.
Rather, our localized approach would break that portfolio down into different markets and first decide for each of them whether there’s any reason low risk shouldn’t work in a particular market. If we conclude it is the right factor to deploy, then we would think about whether there were reasons to modify the way we implemented the low risk factor based on something unique to the market we’re considering.
Let’s stick with low risk as an example. That turns out to be a signal we think should intuitively apply in all emerging markets. We think retail investors generally have a preference for gambling, and so they tend to pay too much for risky stocks and overlook stable, low-risk companies.
As with most of our signals, we try to capture a feature like low risk in multiple ways. One approach to scoring stocks on riskiness is something we call our “lottery” signal. This is basically flagging those stocks that had the biggest one-day returns in recent months. We understand that these big payouts attract the attention of retail investors who rush out to buy stocks that feel like financial lottery tickets, which means they tend to bid those very risky stocks up to irrationally high prices. Our strategy is to underweight these overpriced lottery stocks.
Now, in China, one thing you’ll find if you study the way trading is regulated is that the exchanges impose daily price limits on individual stocks. If a stock goes up or down by more than 10% in a single day, trading’s effectively halted until the next day. So the maximum return you’ll see is 10%. That means if you ranked stocks in terms of their biggest one-day returns, you’d have a whole bunch of companies tied at 10% and the traditional lottery signal wouldn’t differentiate among those.
Once you recognize how this particular aspect of market structure affects the model, one way of enhancing the lottery signal in China would be to look into that basket of stocks with 10% max returns and award a more negative rank on the signal to stocks that had multiple consecutive “limit up” days in a month. Those stocks would have received the highest levels of retail attention, and so they’d presumably be the most overpriced.
That’s an illustration of the general approach to localizing in EM: finding things that work in specific markets and using a deep local market knowledge to make the models work even better.
That’s right. We just touched on two forms of localization, both related to how we apply what you might call “standard” quant signals to individual markets.
The example we were just discussing is one means of localizing a standard factor: taking something like the low risk and tweaking it due to some market-specific feature.
I also mentioned before that we’ll usually think about whether there’s a reason to exclude a traditional signal like low risk from being applied to a given market. That’s not something we do very often—to avoid data mining, we like to have strong empirical evidence and also a very strong theoretical reason to exclude a signal—but sometimes it makes sense to drop a signal somewhere.
For example, across most markets, we like to score companies on their tendency to over-invest, which is a decidedly negative feature in most markets. In China, it turns out the story’s a little more complicated. For state-owned companies, very high levels of plowback tend to be a bad sign, which maybe isn’t too surprising. But for non-SOEs, it turns out that in a rapidly growing economy like China’s, where access to external capital is often a challenge, high levels of reinvestment can be a signal of strong growth opportunities, so we don’t apply over-investment as a negative signal among non-SOEs in China.
On top of standard signals that might not apply in some regions or might need to be adjusted in other regions, there are country-specific signals that we simply wouldn’t be able to calculate in most places. But because of some unique feature of a particular market or because we’ve found an interesting data set that applies just to one market, we can come up with a totally new signal that doesn’t even exist among the so-called “standard” factors.
To share one example of that, in China we’re able to get data on the trading activity of foreign institutional investors accessing mainland Chinese stocks through the Northbound Stock Connect channel. We’ve done a significant amount of research on the performance of these offshore pros (see Research Note on “Searching for the Smart Money in China A Shares”), and we find that their trades tend to be quite profitable. Now, since we observe a list of the stocks they’re buying and selling in real time, we’re able to implement a signal that effectively “piggybacks” on the information in their trades. Stock Connect is one of those market-specific features that creates some cool data we’re able to incorporate in our models.
I just gave an example of the benefits of data in a localized framework. Each of these markets we research is different, and so we get to work with some very cool data sets. But there are also some challenges to localization when it comes to data.
Standard quant research makes heavy use of the typical market data—like prices and volume—and, of course, also accounting data from companies’ financial statements. To do localized research, we need to supplement that with data from local sources. Some simple examples would be data on state ownership in China or information about business group membership in South Korea. We can’t get that kind of stuff from the usual global vendors, so we need to find that from domestic sources, either data we can buy from vendors or—even better—data that take some work to extract from sources that might be “under the radar”, so to speak, from the perspective of other quants.
Beyond just getting our hands on the local data, another challenge with localization is data quality. Data are the building blocks for all our research and strategies, so we need to make sure we’re working with solid inputs; otherwise it’s “garbage in, garbage out”.
With local data, it can take some time to fully understand what we’re looking at. That involves not only knowing how accounting standards differ across countries, but also understanding companies’ business practices in different regions. Let’s take trade receivables in China as an example. If you didn’t realize Accounts Receivable and Notes Receivable are often used interchangeably by Chinese companies but were historically reported separately on Chinese financial statements, you might end up underestimating Chinese firms’ receivables when calculating a signal like one of those we use to capture earnings management in China.
There’s also the question of data cleaning. Scrubbing the data thoroughly requires some effort before they’re added to our databases where they’re available for use in our research or in implementing our strategies. Usually the data we get from local vendors can’t just be used “off-the-shelf” without being carefully checked. Even with global vendors’ data, there’s a good deal of cleaning required, and we sometimes need to “undo” the standardization they make so we can get back to the raw data reported by companies in each country.
All of this can be tedious work, but we view it as a critical part of the process of localizing our strategies. I’m convinced spending so much time curating data gives us a substantial edge relative to investors taking a laid-back approach to the data. That’s especially true in emerging markets.
Speaking of which, one other challenge that’s harder to overcome in emerging markets is the fact that in most of these countries, while there are so many interesting facets of the data we can study, we often don’t have a very long history of data. Quant research is really built on the assumption that we’ve got a big enough sample to draw conclusions about patterns and structure in the data, so that can be a big hurdle in building EM quant models in general—and it doesn’t just affect localized research.
Probably the most important thing is not solely relying on data to tell us which strategies to pursue. We obviously like to see something perform well in a backtest, but it’s more important that we have a compelling theory for why something should work—like my earlier reference to retail investors having a propensity to gamble, which pushes the price of lottery stocks too high.
Another strategy for overcoming a lack of long histories of data, where possible, is to test the same signal in multiple markets. So going back to the case of daily price limits in China, it turns out there’s an almost identical mechanism in Taiwan, so that gives us another place to test the enhanced model, and if something works in multiple markets, even if our sample is short in each market, we’d be more inclined to believe the effect was real—especially when there’s a good story in terms of the economics and investor behavior to back it up.
Absolutely. You’ll find that each market is different, but there are these pockets of similarity that lead to subsets of markets offering similar opportunities. Let me offer another example. In China, there are mainland-listed A shares, but a large number of companies, in addition to the A shares, also list H shares listed in Hong Kong. Those two share classes have the same voting rights, the same cash flow rights—for all intents and purposes, they’re identical, except for where they trade.
But because they trade in relatively disconnected markets, the prices of a company’s A and H shares can deviate massively, with one share class sometimes trading at two to three times the price of the other. Because it’s essentially impossible to short sell the A shares, there’s no real way to arbitrage this difference. Still, what we’ve found is that you can use the price ratio of the two share classes as a signal of valuation; if the A share is much more expensive than the H share, it’s more likely to be overvalued, and you probably want to underweight it.
As a side note, that’s an example of a completely new localized signal that uses pretty boring data on prices, which everyone has, to trade on a feature that’s fairly unique to Chinese stocks.
Now, I say it’s fairly unique to Chinese stocks, because it turns out that a decent number of Brazilian companies also have two share classes: ordinary and preferred shares. This isn’t exactly like China’s A-H dual listings. The Brazilian shares trade in the same place yet have some different features in terms of dividend yield and voting rights. But, just like in China, we find we can use the price ratio of these Brazilian dual-class stocks as an effective signal. This is a case of a localized signal—what I might call dual-class mispricings—that we can exploit in multiple places.
It’s actually very fortunate for us that we discover similarities like this, because it means when we come up with new ideas in China or Taiwan, for example, those insights might drive improvements to our models in other regions. So even though we’re doing deep, time-intensive research in individual markets, the occasional overlap gives us that much more bang for our buck.
That’s a good question. First, let’s put emerging markets as a whole into perspective.
At the end of August 2019, there was a grand total of 26 countries in the MSCI EM Index, with China accounting for the greatest weight at around one-third of the portfolio. The top five markets—China, South Korea, Taiwan, India, and Brazil—made up just over 70% of the index.
It’s probably clear from our conversation so far that localized research isn’t easy. It takes massive effort to obtain and clean the data, and to learn enough about each market’s features so we can pick things apart and create distinct models for capturing those local differences. So far, we’ve focused mainly on the benefits of localization and touched on some challenges, but I should note that it’s also a pretty costly endeavor from a research perspective. So we have to be conscious about the costs and tradeoff we face when we decide to drill down and deeply understand one region.
To prioritize coverage, I like to think of allocating research resources roughly in proportion to each country’s weight within emerging markets as a whole. The advantage of that approach is when we come up with a market-specific insight, we can be sure it impacts enough of the portfolio’s weight to make a meaningful difference in the overall performance. We still apply standard factors everywhere, and we’re constantly doing research to enhance those standard factors—basically coming up with a proprietary set of non-market-specific signals. We know that’s worthwhile because it touches every region. We also get the benefits I talked about a moment ago; where we learn something in China, for example, it can be generalized and applied to a handful of other emerging markets. That does help a lot.
The thing about emerging markets is that they’re still in the midst of this life cycle that will eventually, hopefully, see them converge to developed market status. But they’re starting at different times and places, so there are so many unique features that create opportunities for localization. On top of that, emerging markets typically show much higher levels of retail investor participation—which we’ve discussed in a previous Research Note (see “Where Retail Rules: Buying in China’s Alpha Opportunity”)—and that leads to more mispricing and perhaps also a greater payoff to quants who specialize in EM.
Developed markets are more mature, in terms of both having a more sophisticated investor base—so less alpha to exploit—and having already converged to more homogeneous features in terms of accounting standards, regulations, market structure, and so forth.
Of course, all markets are complicated, so I think there’s always an opening for doing highly specialized research on particular features of a given market—even developed markets. But for the reasons I just mentioned, the conditions in EM are probably just more fertile for localization right now.
Generally speaking, quantamental investing is this idea of somehow blending fundamental research with quant research to get the best of both worlds. We could have an entirely separate conversation about Rayliant’s quantamental approach—and I think we should—but let me just briefly touch on how fundamental research comes into play in the work we’ve been discussing.
First and foremost, we are not fundamental stock pickers, but rather quant investors designing systematic investment strategies. But we recognize that the best quant models are built around a strong understanding of economics, human behavior and the real-world business underlying our data. The most effective way for us to capture that and build the best models is not to spend all day staring at numbers on a computer screen—we might never figure things out. It is instead to talk with fundamental researchers who have local knowledge in particular markets and learn what they know about the features of these markets and data sets we might be able to use.
So for Rayliant, fundamental research—and this quantamental mindset—is critical to making our models truly localized. In fact, the earlier examples of daily price limits and A-H dual listings are both features our fundamental research team told us might be of interest, and that led us to gather the data and figure out how to incorporate those elements into a systematic model. It’s an exciting process, to see the fundamental and quant teams working together in that way.
We view localization as extremely important if you want to harvest the most alpha in EM, and it adds significantly to performance, but we still capture plenty of mispricings through standard quant signals. Even something as simple as sorting companies on a value indicator, like book-to-market ratio, is going to work very well in highly inefficient markets where you’ve got droves of retail investors constantly trading to push prices away from fundamentals.
As mentioned earlier, we also spend a good deal of time researching the non-localized elements of our strategies. But we do feel there’s a large amount of alpha left on the table if we overlook local insights in various emerging markets. Indeed, a large part of the puzzle that we’re trying to solve when we look at a company through the lens of a quant model is missing if we ignore market-specific, localized elements like those we’ve been discussing.
Subscribe to receive the latest Rayliant research, product updates, media and events.
Issued by Rayliant Investment Research d/b/a Rayliant Asset Management (“Rayliant”). Unless stated otherwise, all names, trademarks and logos used in this material are the intellectual property of Rayliant.
This document is for information purposes only. It is not a recommendation to buy or sell any financial instrument and should not be construed as an investment advice. Any securities, sectors or countries mentioned herein are for illustration purposes only. Investments involves risk. The value of your investments may fall as well as rise and you may not get back your initial investment. Performance data quoted represents past performance and is not indicative of future results. While reasonable care has been taken to ensure the accuracy of the information, Rayliant does not give any warranty or representation, expressed or implied, and expressly disclaims liability for any errors and omissions. Information and opinions may be subject to change without notice. Rayliant accepts no liability for any loss, indirect or consequential damages, arising from the use of or reliance on this document.
Hypothetical, back-tested performance results have many inherent limitations. Unlike the results shown in an actual performance record, hypothetical results do not represent actual trading. Also, because these trades have not actually been executed, these results may have under- or over- compensated for the impact, if any, of certain market factors, such as lack of liquidity. Simulated or hypothetical results in general are also subject to the fact that they are designed with the benefit of hindsight. No representation is being made that any account will or is likely to achieve profits or losses similar to those shown. In fact, there are frequently sharp differences between hypothetical performance results and the actual results subsequently achieved by any investment manager.