Uncovering the Hidden Risks in Your Portfolio

Audio Description

Is your portfolio riskier than anticipated? Sometimes dissociated stocks move together and expected diversification fails. Cluster analysis can help uncover these unknown connections. Guests Nelson Yu and Peter Chocian from Bernstein’s Equity Quantitative Research team join Matt to discuss how to locate and combat these risks.

Transcript

00:00 - 00:15

There are mounting risks in the global financial markets, whether it's the effect from the continuation of trade tensions or the impact of negative interest rates across the globe. Investors today need an expanding set of tools to help them understand the investing landscape and make well-informed decisions.

00:16 - 00:51

Cluster analysis is one such tool. Hi, everybody, and welcome to The Pulse, where we cover trends in the economy, markets, and asset allocation for long-term investors. I'm Matt Palazzolo and today I'm joined by Peter Chocian, senior quantitative analyst and portfolio manager on the phone from our UK office. And Nelson Yu, head of equity quantitative research here in New York with me. Peter and Nelson have done a lot of research on how cluster analysis deciphers complex relationships to detect unknown risks lurking in market behavior patterns.

00:51 - 01:01

So, Peter and Nelson, welcome to the show. Hi, Matt. Hi, Matt. Thank you very much. Yeah, sure. So let's start with understanding what Cluster analysis is. What's the history behind it? Maybe Nelson, you start. Sure.

01:01 - 01:35

So, look, when I'm looking at a portfolio on an everyday basis, we're watching the daily movements of the stocks in that portfolio. Right. And what often struck me was that the degree to some of these stocks moved together in ways that we wouldn't have thought. Our traditional tool sets would just tell us that these are unrelated companies. Not correlated with one another. That's right. And so you might have a company media and a company retail just moving up and down together. And that's not something that, you know, as a portfolio manager, we intended to have. And that movement up and down is more than would be implied by just the stock market being up that day or down that day. That's right.

01:35 - 01:56

The correlation is even tighter than would be implied by that. Yeah. And so that really puzzled us in terms of what was going on. And, you know, when you think about a traditional risk model, right, it just says, media and retail are pretty uncorrelated. It's driven by different factors. That's right. Yeah. Media should be driven by, you know, ad revenues. Retail should be driven by consumer spending.

01:56 - 02:15

Right. Right. And so that is a problem. You know, we're trying to put together a portfolio and really bring about that diversification by having these individual stocks move differently. So essentially, that diversification that you thought that you had doesn't actually exist because these stocks are moving together. And I guess what you're going to say is that's the clustering. That's right. Yeah.

02:15 - 02:48

And so the problem that you have as a portfolio manager is when these stocks start moving together in ways you weren't expecting. If a market events happens, these will draw down simultaneously. And that's really going to create a lot more risk in your portfolio than you had actually anticipated. So, look, I remember discussing this with Peter almost a decade ago, and we were just talking about what are different ways that we could actually be addressing this issue. So, Peter, maybe walk us through that conversation. You were familiar at that point in time with some of the research around these issues. So maybe let our listeners know how that folded into investment management research.

02:49 - 03:19

What it sounded like Nelson really wanted to do was to find a technique or method to be able to understand why these stocks were moving together, but moving in different groups than the risk models were placing them in. So one technique that I was familiar with was cluster analysis. So it was fortunate during our discussions that I could appoint Nelson in a direction which I'd already done research and that could then help us.

03:20 - 03:40

Cluster analysis actually goes back a long way. It's got a pretty long history. It began in 32, 1932, in a paper called Quantitative Expression of Cultural Relationships. So an anthropological study. After that, its progress was slowed, mainly because it's quite a computationally difficult thing to do.

03:40 - 04:11

And it wasn't until the 60s when computer power started to increase, that it became more commonly used then in a whole range of fields, including biology, where finding categories for organisms and cells is really important, but also for other things like political campaigning, where finding characteristics of groups that are likely to support candidates or issues is a really important and useful thing to do. It's been used in medicine, for example, for the segmentation of MRI data, and economics, psychology, geography, literature.

04:11 - 04:31

In fact, anywhere where uncovering relationships can help generate insights, which, as you can imagine, is pretty much every field. Let me just jump in here. I think this is fascinating because what essentially you're saying is this clustering analysis, which is now being used to help portfolio managers manage their risk, didn't even start in finance.

04:31 - 04:45

It started in anthropology. It was used in biology, psychology, polling, it sounds like. And that's fascinating. Absolutely. It has been used on and off academic financial papers, but never really in the industry.

04:45 - 05:12

The last few years on the sell side, it started to see some use, for example, looking at how factor returns have been moving together. So understanding which group of factors work together has value and lower beta stocks. Have they been moving together or have lower beta stocks and growth stocks been moving together? So a whole bunch of insights have been calculated from the sell side recently, but not really from the buy side.

05:12 - 05:30

And the issue there is the complexity. When you've got a market that might be 1,500 stocks, it's a lot harder than looking at, for example, 10 or 20 factors, and now they're moving together so quickly. Yes, yes, so it's effectively our computing technology has caught up.

05:30 - 05:40

That's great. Yeah. So one of the things I think that's been really valuable with having Peter on the team is just he's got a very different background. He came to us from meteorology. Wow. Yeah. We should have started there. Maybe.

05:40 - 06:05

Peter, what's your background and how did you get into this industry? My background is theoretical physics. So several years doing as a postdoc in atomic physics and then a few years in meteorology. So anything where there's a numerical analysis and simulation involved, that was my game. And finance actually seemed a natural stepping stone.

06:05 - 06:35

I had friends who had made the shift and they said, you know, you can get a lot of thinking done and a lot of work done. And that's exactly what I think happened when we started on this clustering analysis part. I think that's great. Just to show that the background, the varied backgrounds that our analysts have, Peter coming from his very unique background and being able to leverage his prior insights and knowledge in those fields to now understanding whether or not stocks are moving together or should or shouldn't be. That's right. Yeah.

06:35 - 07:09

It really, you know, talks about how you can pull in different disciplines in different ways of analyzing things. And I think with finance, you know, it's, we haven't really applied the science yet. And that really opens up some doors here. So I want to pull back to these risk models, clustering analysis being one of them. But, Nelson, you help our portfolio managers across the firm with looking at different risk models, different ways of assessing whether or not their portfolios have risk and if they do, where are those risks concentrated. Describe those models for our listeners, if you would. Sure.

07:09 - 07:35

And look, I think there are probably about four categories of how you have to think about risk. OK, the first category is really the most important where we spend most of, the most time. And that's the risk of individual stocks, right, the idiosyncratic risk. I think that probably what most listeners think about when they think about risk is Procter & Gamble or Caterpillar. Any of these stocks, are they risky, in and of themselves? Yeah. So it's really a question of, OK, what do I think Procter & Gamble is going to do on their earnings announcements? Are they going to hit beat or are they going to miss?

07:35 - 07:35

Right.

07:36 - 08:11

And so we can attack that from both the quantitative or fundamental way. Quantitatively, what we can do is we could look at the history of their stock price patterns and we can actually evaluate, well, how much of that tend to move around on a day-to-day basis. And we're often doing both of those. Right. We're doing the quantitative research and the fundamental. And we're marrying those two. That's right. Yeah, with the fundamental, you're looking at what's the range of outcomes that are likely, and what are going to be the drivers of that earnings outcome. So one of the risk models is obviously idiosyncratic, maybe. What are the others? So then we have our traditional toolset, which is you'll hear talked about as a factor risk model. Right.

08:11 - 08:34

And so the problem with evaluating risk and I just mentioned Procter & Gamble and you would look at its daily history, but you actually have to make a lot of estimations, right. So if you think about the US large cap market, it's close to a thousand stocks. And so if you think about those thousand stocks, you need to not only measure how much each individual stock move around, but you'll also have to measure how much the stocks will move together.

08:35 - 09:00

Right. That correlation that we always talk about. And so that's close to a million different estimates that you have to make across, you know, these pairs. And so how do we go about doing that? There's really not enough data to be able to make, you know, those, that number of estimates that you need. And so what we do in risk modeling is we actually simplify it down and we look at, well, what are the known sources of risk, right?

09:00 - 09:16

What's the geography of the company? What's the sector of the company? Is it large, is it small? Does it actually, is it selling cheap? Is it, does it have high profitability? Is it leveraged? These are all the characteristics of individual companies or stocks that makes them move together.

09:16 - 09:39

That's right. That would be those factors. Yeah. And so, you know, using that analysis, we can have a pretty decent estimate of, over long periods of time, how much stocks will move and who they'll move with. OK, so that's another set of tools that we use. Then, you know, being that we know that we experience the portfolio on a day-to-day basis and not on a long-term basis,

09:40 - 09:55

right, we'll want to know, well, is this portfolio going to be sensitive to macro events? Right. So let's say oil prices actually go up by 20 percent in a given year or come down by 25 percent. How is that portfolio going to behave? Right.

09:55 - 10:20

Because you've got things like airlines, you've got oil companies, you've got automobiles that are oil sensitive in different ways, and then you've got technology that really shouldn't behave with oil. So movements in oil prices is just one example of a macro event that you can run through the models to see what's the impact on individual companies, I guess, movements in interest rates and other issues or other macro events that you run through the model to see what the impact is. That's right.

10:20 - 10:49

Yeah, we'll test things like rates, oil prices, currency is something that we test, we look at commodity prices, these are all things that will run through scenario analysis on the portfolio and just make sure, are we sure that we're not making a bet on something that we really don't have a view on. So how is then Cluster analysis the same or different than these traditional risk models? Is this a replacement to what you guys are already doing or does it augment what you're doing? No, it's a supplement.

10:50 - 11:20

And what I would say is, if you think about what I just laid out here, right. The thing is, all these risk models start with, what do I know? Right. I knew to ask about oil prices. I knew to ask about, you know, interest rate sensitivity. I knew about sectors and geographies. What I'm trying to do is just measure that sensitivity to things I knew. With cluster analysis, it's different. It's challenging me to actually discover new risks. Right. What are the patterns that are moving together that maybe I hadn't thought about? Right.

11:20 - 11:21

If you hadn't identified it,

11:21 - 11:55

but these stocks are moving together, that maybe opens your eyes to areas that you need to do more research to try to understand why they're moving together. Is that fair? That's right. And that's really, you know, one of the tenets of risk management is you really have to keep pushing and probing yourself for what's in the portfolio. So, Peter, it may be helpful for our listeners to get an example, something a little bit more tangible to help them understand clustering analysis. Do you have anything like that? Sure. What we do in portfolio meetings is, as Nelson says, we use cluster analysis of the framework.

11:55 - 11:58

And what it helps us do is to ask questions.

11:58 - 12:23

It makes us question the various states and results we have from the other risk models and sensitivity analysis and the factor characteristics of stocks. We have all this data and then cluster analysis acts as a framework. So what it will do then, it will make groups of stocks that have moved together. And the goal of us in the portfolio review team and portfolio management team is to understand why the stocks have moved together.

12:24 - 12:49

So part of that understanding is going through all the other data and asking a lot of questions. And what we end up with then as we end up with the clusters that have been identified with the calculation, and then we look through them and see how they differ from standard regional and sector and industry memberships and whether we're over- or underexposed to them versus our benchmark.

12:49 - 13:02

So what clusters have you seen recently in some of these research reviews? What has popped up? Sure, in a recent meeting, we were looking at developed-world clustering and this was mid to large cap, just to define our universe.

13:02 - 13:32

And previously, a few quarters ago, what you'd see mostly is that regions were dominating. So USA stocks would be all in a group, and then European in a group, Japanese in a group. And those would be the dominating structures. And within those regional groups, you might get differences. So you might get the USA financials, and then next to them some USA tech and cyclicals. But what we found over the last few quarters is that it's the risk of the stocks in terms of the systematic risk. So the beta.

13:33 - 13:53

So what we've seen is a split between high beta cyclicals such as financials, consumer cyclicals, [...] and lower beta defensive, such as consumer staples and utilities and telecoms. And within those groups now, the regions are all mixed up. So we're not seeing regions dominate.

13:53 - 14:23

But this cyclicality risk. And what we've noticed as well within those large risk clusters is, for example, that in the low beta defensive stocks, as interest rates started to change, that the bond proxy, so the utilities, telecoms, and consumer staples, split apart from the other low beta defensives. So you are seeing that interest rate sensitivity was driving some of these stocks as well.

14:23 - 14:38

And even though you could classify them all as defensive, these were behaving very differently. So whether we were over on the way to this cluster mattered a lot more than we might have understood without doing cluster analysis.

14:38 - 14:40

So, Matt, just imagine this.

14:40 - 15:00

So if you were a portfolio manager and you thought that you actually weren't taking on much macro risk and you had your cyclical autos and airliners on one side. Right. And then you said, you know what, I'm going to put in a bunch of utilities and REITs on the other side, OK? And that was my defensive positioning. Right. What might surprise you is maybe they didn't protect.

15:01 - 15:21

Right, because you're saying these defensive companies, utilities or REITs or whatever it is that you thought had no economic sensitivity, actually were because when interest rates started to move meaningfully, they sold off and they moved in a cluster. That's right. Yeah. And we saw that back in 2018. Right. OK, Nelson, I know you and I

15:21 - 15:48

have talked about the FAANGs, you know, Facebook, Amazon, Apple, Netflix, Google, as an example of how clustering can really have an impact, maybe walk our listeners through that, because I think there, too, it's also very tangible and underscores the value that clustering analysis can provide. Look, when you say FAANGs, right for people who have been reading The Wall Street Journal, they have a pretty good understanding that that's a group of stocks. Right.

15:48 - 16:16

These are technological stocks that have been disrupting traditional, you know, brick and mortar type of business models. But then if you actually think about how these companies are classified, you've got Google and Netflix in media, you've got Apple in technology, you've got Amazon in retail. And those are three pretty distinct sectors of the economy. So if you think about a traditional risk framework, they would say, these four stocks are really not related to each other.

16:16 - 16:26

They're diversified. That's right. So you could say I've got a lot of consumer exposure, but I've got it, you know, through a pretty wide array of ways to get that exposure.

16:26 - 16:42

I guess you're going to tell me that's not true. No. I mean, and we've seen that they actually will move along with the market to a very high degree. So why is that? Well, look, I think there's a few things. One is, first of all, just the, how people are thinking about it. Right. Fundamentally, you can understand why they move together. Right.

16:42 - 17:14

It's just something that's hard to model. But then I think there's also technology that's at play here, because this is a basket of stocks that someone wrote down and penned and came up with a pretty clever name to it. Right. And so it captured the imagination of the market. And so what they did was they created a basket to it. OK, these baskets, because of the advent of ETFs and electronic trading, are just much, much easier to trade now that you can be bought and sold intra-day. And they are, to big size.

17:14 - 17:41

That's right. And so when they're bought and sold, what's happening is you actually have to buy all four or sell all four. And that causes co-movement of these stocks. So they're all moving together. So that's a cluster that essentially, according to traditional risk models, wouldn't be picked up because you think you've got a retail or a tech company and so forth. That's right. Yeah. So and... if you were just to follow that traditional risk framework, you would say, hey, I like all four of these companies.

17:41 - 18:13

It's across a pretty broad swath of the consumer segment. I'm just going to go and buy them. But what you don't realize is that these are all going to be moving together. And when they move down, your portfolio is going to move down in much more significant way. OK, let's start to bring this all together. So, Peter, I want to come back to you. How should or how does a portfolio manager use all of this information? Let's just say a clustering analysis is done on that portfolio manager's portfolio. Once there's a theme that has developed, what's the next step that the portfolio management team does?

18:13 - 18:43

Well, the main question they ask is, is it intentional, their exposures to those clusters? So what you'll do is you compare your holdings, right, versus your benchmark across the clusters and you look at the ones where the weights are very different, so you might be on the way to overweight. And then you ask, is that intentional? So does the fundamental conviction in your holdings in those classes justify the positioning? The stocks might have done very well.

18:43 - 19:16

And you're very overweight a cluster. So it might be time to trim your positioning. If you're very underweight a cluster, perhaps your analysts should focus, therefore, for new ideas. So the main things that we want to be sure of, with all the questions that have been brought up by the clustering and the work done to identify reasons for the clusters is, are you happy with your positioning? I think this raises a good point. And Nelson, I know you want to jump in here, but being a portfolio manager and being an investor is all about a mosaic.

19:16 - 19:46

Right? You have to understand your portfolio, risks, opportunities from a lot of different angles. And it sounds like clustering is just one extra input into that analysis. That's right. It's, as I've said before, it's a discovery process. Right. And, you know, with good risk management techniques, what you're trying to discover is how might I be hurt by things moving together that I wasn't expecting. That's right. That your traditional models weren't picking up on or just common sense wouldn't identify as something to look for. That's right. OK, we're going to stop there.

19:46 - 19:51

And I thank you guys for joining me. Nelson and Peter, we appreciate your insights. Thank you very much, Matt.

19:52 - 20:20

Thank you, Matt. And thanks to all of you for listening. If you'd like to learn more about our insights on the capital markets, see the link to our blogs in this episode's description. And if you enjoyed this episode and haven't yet subscribed to our podcast, please go to the iTunes store, or Google Play, or wherever you listen to podcasts to subscribe and to rate us. And finally, please e-mail us with your thoughts or questions or feedback to insights@Bernstein.com and be sure to find us on Twitter at BernsteinPWM. Thanks, everybody, for listening.

20:22 - 20:30

Bernstein: Making money meaningful for individuals, families, and foundations for over 50 years. Visit us at Bernstein.com.

Hosts
Matthew D. Palazzolo
Senior Investment Strategist—National Director, Investment Insights

The information presented and opinions expressed are solely the views of the podcast host commentator and their guest speaker(s). AllianceBernstein L.P. or its affiliates makes no representations or warranties concerning the accuracy of any data. There is no guarantee that any projection, forecast or opinion in this material will be realized. Past performance does not guarantee future results. The views expressed here may change at any time after the date of this podcast. This podcast is for informational purposes only and does not constitute investment advice. AllianceBernstein L.P. does not provide tax, legal or accounting advice. It does not take an investor’s personal investment objectives or financial situation into account; investors should discuss their individual circumstances with appropriate professionals before making any decisions. This information should not be construed as sales or marketing material or an offer or solicitation for the purchase or sale of any financial instrument, product or service sponsored by AllianceBernstein or its affiliates.

Related

Update browser for the best experience

We may not support your browser anymore. For the best experience, we recommend using the most recent version, or one of our supported browsers.