Vanguard's Return Forecasts Explained: What the Percentiles Really Mean with Kevin DiCiurcio (EP.241)

0:28We all need to make smart decisions with our money. expected return, the volatility, and the correlations that drive our financial planning models. And I also explained how a number of other people approach that activity. Now, one of the ways that some people do it is adopting institutional assumptions. And today, I have Kevin DeCercio, head of Vanguard's Capital Market Model Development. He's going to talk about the process they use to set up those models and develop the outlook that a wide range of people use. Now, I personally like using these capital market assumptions for setting expectations.

1:03If you go back and listen to the last episode, you'll see how we do it at PlanCorp. But Kevin was actually really, really important to evaluating that process a few years ago and thinking through the different solutions that were available to us, the way that we had been doing it. And we do touch on some of those things in this conversation. But this is a pretty wonky conversation, I must warn you, but I do think it has a lot of value. And one detail that kind of sticks out to me, and I'm recording this right after we got off of the conversation. Every time I thought about how much effort they are putting into their models.

1:36My goodness, think about all the people on Wall Street or in all financial markets across the world who are putting so much information into their buy sell decisions. And I'm not saying markets are perfectly efficient, but the competition in markets does make it blatantly obvious to me, at least, that if you think you have some sort of gut feeling or intuition that makes you want to do something to your portfolio based on the information you have, just listen to a conversation like this and see how much thought Kevin and his team put into return assumptions. Because chances are there is already some sort of probabilistic consideration for what has you worried or what has you excited about a certain return or asset class.

2:20As always, you can find detailed show notes at the long-term investor.com. And at the top of this episode description, there is a link to sign up to my newsletter, which comes out every other Wednesday in links to resources that I am currently reading, as well as gives you access to exclusive content that you cannot find on my website or on this podcast. All right, let's dive in. Here is my conversation with Vanguard's Kevin DeSersio.

2:50Welcome back to The Long-Term Investor. Today, I'm thrilled to be joined by Vanguard's Kevin DeSersio, who has a very long title, Kevin. We've been talking about this in advance, but when I asked you to come on the show, the real reason that I wanted to speak with you in front of our audience is that you're in charge of the capital markets research and helping develop those models that put out the return assumptions. Many advisors like myself, and I suspect many individual investors see at the beginning and the middle of the year. So Kevin, welcome to the show and tell the audience, maybe in a better job than I have, exactly what it is that you are doing at Vanguard.

3:25Hi, Peter, and thank you so much for the opportunity to be on your show today. And as you mentioned, my name is Kevin DeCercio and I am a research team lead leading a global research effort to inform on our capital markets outlook in our investment strategy group. And our primary, our bread and butter, of our research is the research and development and operation that we've run, which is the Vanguard Capital Markets Model, which is a financial simulation engine that is projecting global asset return distributions to help inform on economic and market outlook, as well as understand portfolio construction implications.

4:02So this is an important part of the research we do, but we also have a broader strategist function within our group, which is really about making sense of what's going on in markets. And we may not be able to predict what's happening tomorrow, but I think a lot of our job is an attribution of an understanding of what has happened to help our investors maintain discipline with their long-term investment approaches. When I reached out to you, you and I spoke, we were trying to figure this out, I don't know, maybe four or five years ago. And Plain Corp, we update our capital market assumptions every year.

4:34Actually, the episode just before this one goes exactly through that process. But while the process doesn't change that much. We do go a deep dive into the process itself and make sure that the process we're using is valid. But let me look at Vanguard's. I reached out right when I saw the numbers come out. Maybe you could share a little bit with us. How do you develop these assumptions, these return assumptions? And how is it that you are hoping that people use them? Yeah, absolutely. I thought maybe I'd start by setting some context and definitions. I don't want to lose your audience at the start with using any kind of insider vernacular, but you'll probably hear me refer to the model as VCMM throughout our conversation today.

5:14Acronyms are very widespread at Vanguard, but it pretty much rolls off the tongue at this point. But VCMM stands for the Vanguard Capital Markets Model. And as I previously mentioned, it's a global asset return forecasting simulation engine. It's really become an integral tool for outlook and portfolio construction considerations that are used across the Vanguard enterprise. Our capital markets research team sits in the investment strategy group, and we own the research development and forecast operation of the model. We're also the gatekeeper of the model use. So we advocate and enable clients to use VCMM in primarily three ways.

5:49First is really encouraging clients and advisors to set reasonable asset and portfolio return expectations. Investors have myriad accumulation goals, most importantly, retirement for most, but also college savings, you know, other general savings goals. And being able to set reasonable return expectations is important information to really facilitate the success of saving and investing. So in the VCMM model, from a high level concept, we're leaning on statistical predictability to set these return expectations. This concept of predictability simply means that our belief about future returns varies through time.

6:24Doesn't mean that asset returns lack uncertainty, just that there's information that exists today that informs on a pattern of returns in the future. I think a useful analogy to describe this is seasonal air temperatures. We may not know exactly what the temperature will be each day throughout the winter, but we do know that winter days on average have a lower temperature than summer days. So similarly, throughout history, intermediate to long-term equity and bond market returns have had quote-unquote seasons, and these can partially be explained by starting market valuations. So therefore, these time-varying returns projected by VCMM give clients a good framework for understanding whether they should consider saving more, spending less, perhaps changing their asset allocation risk posture in order to give themselves a best chance to achieve those accumulation goals.

7:11A second important use case is really VCMM has become an important portfolio analytical tool that helps our clients understand the risk and return trade-offs of their investment decisions. So a first point here really relates to the previous one about setting reasonable return expectations. But because we all invest under uncertainty and ultimately are compensated for that uncertainty, understanding the likelihood of meeting your savings and investing goals really requires a full appreciation of the range of potential outcomes. So VCMM being a globally connected model captures statistical relationships among global asset return drivers, simulates them jointly, meaning we're really trying to capture correlations and higher moments that make this a really important portfolio construction tool.

7:57So this is used in advice settings at Vanguard to quantify probabilities for achieving success, given personal information about savings rates, asset allocation, retirement age, etc. We also have portfolio analytical users who will use VCMM simulations to get insight into how the range of their potential return outcomes or probabilities for downside risk events change based on changes in risk taking in the portfolios. So I would argue that this type of analysis is also useful in revealing preferences about one's own risk tolerance. We can show you a comparison of your projected risk and return profiles between an 80-20 stock bond portfolio and a 60-40, and you may just decide that you're a little bit more comfortable with that range of potential long-term outcomes or the downside risk expectations in the 60-40 to the 80-20, that reveals a risk tolerance.

8:46So analytically, outlook reasons, those are the first two primary use cases. And lastly, what we're using more and more of is VCMM is being used in asset allocation frameworks that will now provide recommendations for the portfolio mix. And really briefly here, we offer valuation-conscious portfolios whose allocations will change through time in order to anchor to either a long-term return target or to help you stay more consistent with that preferred risk and return profile. And then we also perform due diligence on static portfolios and glide paths based on what we call the equilibrium premia assumptions in the model.

9:22Well, I will add one more use case to that, Kevin. I feel like I use it to behaviorally coach my clients to stay the course on their diversifying asset classes, particularly in the last handful of years where U.S. large cap stocks were just beating everything on the planet. And a common question is, well, why do I even own international? And I could often point to Vanguard's 10 and 30 year projections to say, well, look, they're forecasting and there is a range of outcomes, but they're forecasting them to be higher. I am going to get into some of those numbers a little bit later, but I actually would first love to hear a little bit of an insider's look on the process of developing these numbers in the first place.

10:00I appreciate your use case. I think that's a really informed and educated use case of VCOM. I know we spoke a few years ago. So I appreciate the work that you're doing with your clients as we think about using Vanguard or using the capital markets model to benefit a wide do-it-yourself investment audience and an advised audience. So in terms of the process, so we have a 10-person global research team responsible for the research development and operation, as I mentioned. We have governance from internal operational risk teams. Given the widespread nature of its use, we have a lot of governance oversight from operational risk teams, audit teams, as well as internal oversight from our regional and global chief economists, and who are really weighing in with opinions on model methodologies and the forecasts that we're releasing to our clients.

10:47The forecast operation itself is really designed to be as systematic as possible. We re-estimate the model quarterly as new data is finalized and we will run forecasts monthly based on those prevailing market conditions. And we try not to touch it. There is no ad hoc overlays. A lot of that we try to build into the model methodology. But ultimately, we generally release a new long-term outlook each quarter. But in the event of volatile market conditions that are creating big impacts to the forecast, we'll certainly share updates intracurter through our corporate communications teams or strategic communication teams internally.

11:21But ultimately, forecasts will change only as market conditions change, forward-looking information changes, or model methodologies change. So the research development effort really is separate from operating the model on a monthly basis. Any methodology changes or changes to forward-looking assumptions resulting from that research and development effort, that's then part of a broader new model implementation and a change management process. So any model development is only as good as its inputs, but in my understanding, it's a very probabilistic approach. Is that a fair assessment? Yeah. Yeah.

11:53The goal of the model is to capture the full range of potential outcomes. Absolutely. There are so many capital market assumptions that are made by large financial institutions, again, by financial advisors who then need to populate financial planning software with capital market assumptions. Would you mind just quickly touching on maybe some of the different approaches that you did not take that others maybe take in the industry with capital market assumption development? Yeah. Yeah, there's primarily, I think, a couple of ways to do this and then maybe a lot of ways on the how. But a lot of the forecasting really goes back to some of the original Robert Shiller cape work that identified this inverse relationship between starting market valuations and future returns.

12:37So there is this predictive regression approach where you're really just mapping a starting condition to a long term return and visualizing that scatterplot. That's where the predictability came from. But from an explainability perspective, there's that asset pricing aspect to this as well, where any financial security is getting its returns from two sources. You have an income return and you have a price return. So you're starting with this accounting identity. And then from a asset pricing perspective, thinking in terms of discounted cash flows, that income return on an equity side is coming from dividend yield and your price return can be decomposed into its valuation change and then the growth in the fundamental that you're using that valuation.

13:18So a lot of approaches, you see the sum of parts where you're making an informed view, maybe independently on, okay, dividend yield is today 1.5%. So maybe I'm looking at a historical average there, or maybe I'm just locking in the current condition. Valuation change, maybe my expectation is there's some reversion to a historical average there so that you're just computing that impact to return based on some assumption of what valuations are going to do. And then, you know, earnings growth, you see a lot in literature, historical averages. It's hard to find predictability there. We may dispute that a little bit.

13:51But I think you have some combination of these two approaches of predicting based on mapping current conditions to a future return or making some decisions on those sum of parts. So I think where we are is we're taking advantage of this academic literature of predictability, but we want to give that explainability, but we want to model it holistically. And I think that's where maybe the approach is a little bit different is we're taking any ad hoc nature out of views on these sum of parts. And you're limited to these ways of looking at it. And then otherwise, you can just say, look, historical averages.

14:26I throw my hands up. I don't know what returns are going to be. Historically, they've been 10 % in U.S. markets. But I think what we're trying to do is, again, statistically capture predictability, do it in a consistent way that we're not heavy handed making a lot of assumptions in the model while we're running it. And that it is statistically capturing the full range of outcomes. Again, it's not a different way of thinking it, but it's more of a holistic, consistent, systematic model is what we're trying to achieve. Let me zoom out to the end result. And if you're listening to us on the podcast, there's no way for me to really showcase this without saying it.

15:03Although if you're watching us on YouTube, or you're watching us on Cheddar, you go to the long term investor.com and you can get a better look at some of these returns. But what I'm doing right now is I'm looking at the 10 year forecasts. Let's start there. Maybe we'll go to the 30 year and what you showcase are the fifth percentile 25th percentile 50th 75th 95th. So you have your normal distribution, and I'll try not to keep it too wonky for our listeners who don't want to talk stats. But if we're going to look at the 50th percentile outcome, our base case, over the next 10 years, U.S. equities earning an average annualized return of 3.8 percent.

15:38Developed international, you have 6.3 percent. I think, interestingly, you have bonds at 4.3 percent, so outperforming U.S. stocks. When you're looking at these different percentiles, though, if I go out to the 95th percentile, well, yes, equities are in their 95th percentile. They end up beating bonds. But when you showcase those percentiles, Kevin, how is it that you are thinking that people are going to be using those percentiles and thinking about them? Yeah, that's a great question, Peter. And I'm happy to try to shed some light on this. The percentiles are used to describe that full probability distribution, which is generated via a deterministic regression approach and then simulating all of the uncertainty in the model.

16:20We don't have to go deep in that. So what we have here is this full probability distribution function that we're summarizing in these percentiles. There really is this literal interpretation here is like if you're looking at the range of outcomes between the 25th and 75th percentile for US equities, for instance, that has the interpretation of the, there's a 50 % chance of returns at the end of the 10-year horizon falling between that return range. The nuance, I think, in VCMM is that there are time horizon implications, though, to the interpretation of model medians or the model averages, as you've pointed out.

16:53At around the 10-year time horizon for equities, and maybe it's the five-year for broad market or intermediate duration fixed income, we do have this increased confidence in the model's ability to capture realized outcomes somewhere in that middle 50 percentile range. That's why we really define the median returns at those time horizons as return expectations. because of that increasing accuracy in the forecasts. If you were to look at shorter term distributions, which we don't really share publicly because again, we want our investors to use them properly. So we're thinking about the long-term because I know investors are fixated on that 50th percentile.

17:26That's the average, that's the return expectation. But for the more shorter term time horizons, basically what we see in the model is a much more uniform distribution. You're equally likely to get any return between negative 20 % and positive 20 % for US equities over the next year. and therefore we have less confidence in saying, okay, well, the median is a return expectation. It's just not as accurate. So as a research effort, what we try to achieve is really the best forecast accuracy to set those return expectations over the median and long-term. And then over shorter-term time horizons, we're really aiming to ensure that VCMM can capture realized outcomes somewhere in the distributions.

18:01I think the 2022 experience was really instructive for us in this regard. So you probably remember 2022 fondly, the beginning of the year. January 2022 was the beginning of this steep decline in U.S. equity markets over the next nine months or so that coincided with the rapid rise in long-term bond yields as we experienced that inflation spike and Fed signaling an aggressive hiking campaign. This caused simultaneous capital losses across both markets and the notorious spike in the stock bond correlation. Equities were down near 25%. I think bonds were down 15 aggregate bonds in the U.S. market.

18:34So we looked at this, you know, are we capturing this in the model? So a 4060 US stock bond portfolio, and I'm doing a conservative one because, you know, how frequently have bonds had a 15 % drawdown in nine months? I mean, it's almost unprecedented, the aggressive hiking campaign that we saw, but the 4060 US stock bond portfolio was down by 19 % over those nine months. So we went back looking at the forecast produced as of December 2021 conditions. VCMM projected a 19 % drawdown of the 40-60 portfolio, somewhere around the 95th percentile of its max drawdown projections. So in that case, we felt like we've reasonably well captured that outcome.

19:11In other words, with these percentiles, Visamib's designed to set return expectations with the median or the averages of the model from the long term. That's where you're fixating on that 50%. But we really want to fully describe the probabilistic nature of returns over all time horizons. I'll take a moment to explain to people why I think this conversation is interesting, particularly if you are a client of a financial advisor running Monte Carlos, or you are a do-it-yourself investor who's using modeling, is that sometimes you might be worried about something bad happening in the market. And the modeling process takes into account, you know, you mentioned 2022 was in the range of expected outcomes, roughly.

19:53And I think the mistake people make when they're trying to set up return forecasts, particularly when they're for 30 plus years, is they forget, well, if we're going to have these modeling, it doesn't mean that that return you're assuming, assuming you're going to vary it and put a standard deviation on it. it's going to capture downturns of a similar magnitude and frequency as we've experienced in the past. And I'm curious, Kevin, when I click over from the 10-year returns to the 30-year returns, 30-year returns hug the average pretty closely over the long term. So I guess, does the base rate, does the fact that the long-term average, does it get a bigger weighting in the model and thus drag returns closer to their long-term average?

20:32Such an interesting question. The way the model, and I don't know if you have questions on this, but the model really has two important states, right? The first is the predictability that we described. We want to capture the impact of current market conditions on our expectation of the future. And then we let the statistics in the model tell us how long that takes. And we're not as concerned with the path sometimes necessarily, but more it's the, okay, well, how long do these relationships with starting conditions tend to play out? Trend growth rates matter certainly as well in that time horizon.

21:02And we think about the very long term, we do have this equilibrium aspect of the model. And you can consider equilibrium as this is when market conditions would have no impact on our return expectations on the medians. We're looking at sort of the steady state. So you can think of steady state as maybe years 21 through 30 in a 30 year forecast. Maybe you want to go a little bit further and say maybe it's 31 to 40. In that state in the model, initial conditions have decayed off and you're left with these long term premia assumptions. So there's a big body of research where we're looking at that and saying, OK, well, what's our long-term average has been for country equities?

21:39How should we position that in the model? Not going to get into that in this question here. But the relationship between the two matters on time horizon. So our 10-year forecast is almost entirely determined by current market conditions. And then you get to this 30-year horizon. In most normal times, your 30-year horizon really resembles that equilibrium set of assumptions. It's not going to vary that much on it. But depending on how extreme current market conditions are, they can begin to start dragging that higher or lower. I think in the case you have them in front of you, I presume that our 30-year forecasts are drugged down.

22:11I think our steady state assumption is around 8%, which is different than the 10-year 10 % that we've seen historically. There's reasons for that. But we think where current market conditions are today have a big impact on even the 30-year time horizon. That's an interesting point we can pivot to. Over the next 5 to 10 years, what is Vanguard? see as the strongest risk return profiles based on your modeling? So I would encourage, I don't know if you're a reader of the economic and market outlook work, Peter, but I would definitely encourage you and your listeners to read the report that we published in December that'll probably lay out all the numbers, the rationale, economic rationale, and maybe statistical rationale for those numbers as well.

22:51We think long-term, so things tend to move fairly slowly. We're considering bigger picture things, megatrends being one of them. But ultimately, we continue to see strong value in fixed income. Expected equity versus premium, which I often estimate if you're looking at those return tables, by looking at the difference in the median 10-year forecast between broad equities and broad bonds, that is my estimate of expected equity versus premium. It might be different than how some others look at it, but that's based on our forecasts and based on the statistical probability. That's very compressed by historical standards.

23:24So this suggests that over the long term, we do not believe that global stock bond portfolios are really expected to receive adequate compensation for taking on marginal equity risk. And this is particularly related to the U.S. exposures. In that VCMM equilibrium state that I described, where you're not impacted by market conditions, we look at credit risk within the fixed income space. Credit bonds in equilibrium provide really reliable risk premia over the long term. So we can actually make this argument from a strategic equilibrium perspective. You want additional credit bonds in your portfolio.

23:59But given the spread compression that we've seen in both IG and high yield spaces, we just don't think that risk return tradeoff is as attractive today. I would argue that to really capture the benefit from your fixed income exposures, you want to be up in quality, your treasuries, you know, really high grade investment grade bonds, if you're going to be taking that credit risk. From an equity perspective, we again, continue to believe that U.S. value provides more attractive prospects than U.S. growth. And this is even in consideration of different AI scenarios playing out, you know, whether it's an AI-led growth boom or if AI disappoints.

24:34We actually think value performs well relative to growth in both cases. And kind of let me lay that out. In the boom case, we think value outperformance is mostly driven by creative destruction processes in which competition for the AI profits begins to erode the excess profitability that we see today in a handful of firms. Other sectors in this AI boom will start to experience profit growth as well. And then some of these productivity gains really show up in consumer surplus as well. So they're not always materialized in profits in a handful of firms. The broadening out of profitability and the erosion of excess profitability, we think it's a strong case continued for U.S.