Gravity Exists

sunday brunch

A Goals-Based Approach to Optimizing Client Portfolios

July 24, 2022


Sanjiv Das is the William and Janice Terry Professor of Finance at Santa Clara University's Leavey School of Business. Dan Ostrov is a Professor of Applied Mathematics in the Department of Mathematics and Computer Science within Santa Clara University's College of Arts and Sciences. Together, Sanjiv and Dan have performed extensive research on goals-based portfolio optimization for wealth management.

Listen to our conversation:

Read our conversation:

Gravity Exists Hello, everyone. Thanks for joining us today. We're going to be speaking with Sanjiv Das and Dan Ostrov from Santa Clara University. They're going to be telling us about the work that they've been doing in recent years researching portfolio management methodology for wealth managers.

Sanjiv is the William and Janice Terry professor of finance at the Leavey School of Business. Dan is a professor in the Department of Mathematics and Computer Science in Santa Clara University's College of Arts and Sciences. Sanjiv and Dan have worked together extensively and have co-taught Santa Clara's mathematical finance course for 15 years. Last year, they – along with Deep Srivastav and Anand Radhakrishnan at Franklin Templeton, published a paper that received a lot of attention in the industry titled “Dynamic Optimization for Multi Goals Wealth Management.” Sanjiv. Dan. Thank you for taking the time to discuss your work with us today.

Dan Ostrov Thank you for having us.

Sanjiv Das Thanks, Andrew. Happy to be here.

GE Terrific. Why don't we jump right into the heart of your recent work in your paper? You proposed maximizing investor outcomes over multiple individual financial/life goals as an alternative to the Monte Carlo approach that currently serves as the foundation for most wealth management planning. Could you walk us through the basics of this approach that you guys are espousing?

SD Sure. Thanks. Thanks for the question, Andrew. I think that's very well put. The basic idea here was to mathematize and put a framework around goals-based wealth management. We will call it GBWM as an acronym. GBWM has been around for more than two decades. Jean Brunel and Ashvin Chhabra, 20 or 25 years back, started working on this concept, and it jibes really well with the fact that investors usually have goals. They're not interested only in maximizing risk versus return. They are really interested in reaching their life goals and their financial goals. What we did in this work was actually built on previous research we'd done on getting to the optimal, that is, maximizing the probability of reaching your goal, rather than maximizing only the trade-off between risk and return that happens with mean-variance optimization.

The framework that we've built is not a static-optimization framework. It's a dynamic optimization framework that actually maximizes the property of reaching all the different goals that people might have over their life cycle while being on the mean-variance efficient frontier. So, we are completely consistent with previous methods of being optimal, but at the same time, maximizing the probability of reaching goals.

Now, the important distinction between mean-variance optimization, which is a traditional way portfolio allocation is done for retirement planning and for other sorts of portfolio optimization, is that their risk is measured in terms of the variation or the standard deviation or the volatility of the wealth portfolio that's being optimized. In goals-based wealth management, risk is measured in terms of the probability of not meeting your goal. And that's really what investors worry a lot about. Will we be solvent through retirement? Will we be able to meet an upgrade of our car 10 years from now and so forth?

So, these two things are not necessarily compatible with each other, because if I want to minimize risk in the risk/return frameworks, I would just lower the risk of the portfolio. But lowering the risk of the portfolio might actually result in not being able to reach your goals because you aren't taking enough risk with a commensurate return to get there. And so, there is a delicate balance between these two risk concepts, and goals-based management emphasizes that reaching goals is important. So, not getting there is the risk you incur. What we've built in this entire framework was how to do that while not ignoring the fact that there is a risk-return trade between the reach for return and the risk you take in your portfolio as well.

So, that's one of the biggest changes away from traditional wealth management that goals-based wealth management brings in. And our approach sort of deals with this, not just statically, but in a dynamic optimization framework.

The second thing I think Dan should talk about, which is comparing this framework versus a Monte Carlo framework. And that's the second big idea that got implemented in this paper.

DO We are aware that a lot of people have used Monte Carlo techniques to understand how to optimize investment portfolios. But when you do that, you're really trying to think about what is the specific static investment portfolio that will optimize things going forward in time? What we wanted to do was to say: What is the investment portfolio that I should optimally have now, knowing that I can change the investment portfolio in future years, depending on how the market changes and how the wealth of my portfolio changes?

So, if you try doing that sort of thing with Monte Carlo, you almost immediately have something that's computationally completely intractable. That is, you're looking at so many different possible wealths in the future. And then, who knows, maybe 15 different possible investment portfolios for each of those wealths. You don't have enough computational time, even if you ran the computation over the history of the universe's existence, to be able to compute this. So, you really can't go forwards in time the way that Monte Carlo does by nature. What we use instead is dynamic programming. And dynamic programming by its very nature, works backward in time, and it allows us to whip through these sorts of questions in remarkably little time — generally under 15 seconds or around that area. Sometimes it's one or two seconds.

GE In your paper, you compare this active process to buy-and-hold strategies and target-date funds. And what was your process for that comparison? And what were the results you arrived at as far as how this performed versus those?

DO I think this is a great example of where goals-based wealth management really has tremendous advantages. Because when you do goals-based wealth management, you're doing individualized management. You're understanding exactly what these circumstances for the investor would be and how you optimize around those specific circumstances for the individual, as opposed to doing things for the group. Target-date funds are able to take only one thing into account, and that's the age of the person that they're considering.

Let me give an example. Let’s say we have somebody who is 50 years old. They have a hundred thousand dollars. They know they're going to contribute 15,000 present-value dollars per year into their retirement account for the next 15 years. Then, for the 15 years after that, they're going to take out 50,000 present-value dollars each year.

We just want to optimize the probability that they don't go bankrupt. And again, that's very much a goals-based wealth management point of view. Sanjiv was talking about how we minimize the risk from a goals-based wealth management perspective, that is, the risk of not going bankrupt. That's what we're looking to optimize. If you use a target-date fund, you will basically have a 26% chance of not going bankrupt. Using dynamic programming and the methods in our article, that goes up to 63%.

The reason it goes up is that we are able to take into account all of those things. We know exactly how much money we're going to be putting in, and how much money we're intending to take out. And the fact that we are allowing ourselves to move the investment portfolio optimally each year, depending on what the wealth of the investor happens to be.

SD Let me add to that as well. The key improvement -- and it's pretty obvious -- is that you are allowed to update your portfolio to move between high risk and low risk over its life cycle. This movement depends on whether your portfolio performed great in the early years, in which case you can sort of take your foot off the gas a little bit and reduce risk so that you don't end up not meeting your goals by taking too much risk, or, if you haven't been able to get enough appreciation to get to your goals, you can probably add a little bit more gas in our model. Our model computes how much extra speed or risk you want to take to get to those goals.

None of this can happen in a target-date fund because a target-date fund specifically says: If you're 55 years old, or you're 58 years old, this is the stock/bond combination that you're going to live with. It has no cognizance whatsoever as to whether your portfolio up to that point has done well. Are you on track to meet your goals? Are you not on track to meet your goals? It just ignores all that information. So, it's only a function of time. It's not a function of how your portfolio is performing relative to goals at a particular point in time.

We consider a lot more information when making portfolio choices in the goals-based wealth management approach. We weren't surprised that it would do that much better. A 63% chance of a person being on-track to get to their goals versus 26% for a target-date fund was a surprise however, not that the target fund would do worse because it's intuitive as to why it doesn't do as well. But just in terms of how much worse it did was a surprise to us.

DO I think the other thing is that this method allows you to play with any different method for putting infusions in. And so, if you're talking with a client you can say “If you put this much money in for this year and this year, here's how that's going to change your probability of being able to remain solvent.” You now have a really informed conversation knowing that you're optimizing everything that you can within the algorithm. The client knows, “Gee, this is what my situation is if I really want to spend X amount of money in retirement.” Whatever they choose, they have a sense of their likelihood of going bankrupt.

One of the other things that we've done with this research is work that through with the probability of mortality as well. So, you have a very good sense of what your chance of going bankrupt, in your lifetime, is going to be. You can make very informed decisions about how much money you want to save.

SD Just to add to what Dan said, this is just one paper out of a collection of papers that we've been working on for the last — more than five years, I think? Correct? So, this was actually the third paper we started working on, but there's been a few more that account for mortality risk, annuities, and several other bells and whistles on top of the work that you have referenced.

GE Right now, obviously, we are talking to an audience of wealth managers, and many of them are relying on off-the-shelf, management, risk management, and planning software. They are employing standard models. Are you currently planning on providing any kind of model commercially that's going to harness this methodology? Is that something that you'd be doing on your own or in partnership? Do you have any plan for the practical implementation of your ideas on the ground level by wealth managers?

DO Actually, what’s interesting is that we've got a series of papers — I think it's actually up to seven — that we have worked on or are currently working on. All this research has been with Franklin Templeton. So, our two co-authors who are not here today, Anand Radhakrishnan and Deep Srivastav, we've been working with them throughout the entire project. In fact, Franklin Templeton came to us originally with these questions and said, “How can we do goals-based wealth management in a more interesting way? In a deeper way?” And it has led to this just tremendously interesting set of questions and work that has been a combination between Franklin Templeton and the two of us in academia.

So, everything that we've done has always been with the purpose of this being available to investors and them being able to help their clients optimally. It's always been a key part of the optimization work.

GE Does that also mean that you were able to work with live clients as part of the research process and that was able to be incorporated into your research process and the conclusions you drew?

SD Yes. I think that it was 2017 when Franklin Templeton ran a bunch of focus groups at that time. These focus groups essentially found that clients were much more comfortable when spoken to in the language of the probability of reaching your goals versus risk metrics like the standard deviation of the portfolio or the correlations of their portfolio with other things.

Every time there were standard statistical concepts used, the client was very lost. When they were spoken to in terms of the probability of not reaching their goal, or asking them questions like “With what probability would you like to meet your college fund goal of $250,000 in eight years?” the results were very different versus a question like, “What's the standard deviations that you're willing to live with?” They didn't understand those questions sufficiently to be able to give good information to the financial advisor.

GE Right.

DO That's had a huge influence on the way that we've approached the entire project. We want everything as much as possible to be about probabilities of reaching goals so that you can have the conversation with clients in the language they can truly, genuinely understand.

One of the things that has been interesting about the specific paper here is that it deals with multiple goals at once. And when you've got multiple goals, you face a fundamental question which is: How important are each of these goals to you? From an academic point of view, what you do is attach the concept of utility to each of their goals.

The utility just means the importance of the goal, because you want to make sure that you are prioritizing more important goals for clients versus less important goals, but you can't really have them try to attach utility values to that. That doesn't make sense. Do you love somebody? Yes, a lot. Well, how many units of love? That's an impossible question to answer.

So, what we have instead is that we're working through what the probability is that corresponds to what we, behind the scenes, know to be the utilities -- that is, the importance of the goals to the investor -- but all the investor sees is what the probabilities of fulfilling their goals are. Then they can say, “Oh, gee. Of these five goals that I specified, maybe I really want the probability of attaining this goal to be higher.” So, we are now able to adjust things to say, “Okay, this one's higher.” That, of course, is going to lower the probability of the other goals but now the investor is able to really understand the ramifications of their decision strictly by talking in terms of probability.

GE Let me shift the topic slightly. Our audience is comprised almost entirely of wealth managers. Whether they're at independent RIAs or they are at banks. If they are at multi-family offices or single-family offices. Many of them within their practice have institutional accounts. Perhaps endowments, occasionally small private pensions, or various nonprofit clients. You don't really touch on how this might apply to the institutional space directly. Obviously, with everything that's going on right now, the prospect of hitting “CPI plus five” in the coming years seems very daunting. How might this approach be used by institutions?

SD This research has been written from the point of view of an individual investor with individual goals and very much fits the mold of customizing for the individual. One size does not fit all and so on. But the fact is that goals for individuals can also be viewed as liability tranches for institutions.

A pension plan, for example, might be one thing that an institution might be interested in optimizing. You could think of an annual tranche of liabilities as goals in our framework, and we can weight these goals. We might decide that we want to weigh all the recent goals much more and weight the later goals less, so that we at least make the current payments. You know, sacrifice future generations in the plan, so to speak. It's literally the same framework. Liability tranches are goals from the point of view of a pension plan.

The rest of the mathematical framework works just as well. In fact, I think it does better because what it does is recasts the standard pension problem as a goals-based problem. It allows you to back out, using our technology, the probabilities of reaching each and every one of these different lability tranches, rather than just a single number measuring “fundedness.” That is, whether the plan is funded, what is the funded ratio, or how much is it overfunded or underfunded? So, you can extend the ideas for retail investors towards ideas for institutional investors.

DO You mention “CPI plus five.” Our goal here isn't to specify what the investments can do. There is a tremendous amount of disagreement about what capital market expectations are going to be. We want people to come in and say, this is what we think they are and, whatever we are given, we can then plug that into our technology and our algorithm to come up with probabilities given how they perceive the market is going to be in the future.

GE I can see where this could also be very applicable to defined contribution programs if the sponsor wanted to use your methodology, working with plan participants.

DO Right, and defined benefits as well, as Sanjiv was talking about before.

SD The interesting thing also is that, when you think of liability that's, let’s say, 30 years out, and you've got a portfolio that matches that liability 30 years out, when the market isn't performing well like what we are experiencing right now, the portfolio can drop by 15 or 20%. That panics people a lot. If they didn't look at the drop in the portfolio, but instead recalculated new market expectations in our model — what the probability of reaching that 30-year goal is — it may not drop that much. In fact, if you run numerical simulations using the model, it doesn't. The probability of reaching that 30-year goal, let's say, is 98%, and then the market drops. The probability might decline from 98% to 95% or 93%, even though the market dropped 20%, because the strategy dynamically updates itself, and we have 30 years of runway to catch up. Right?

So, in terms of the goal, the probability doesn't drop that much. That's a really nice thing, because people look at it and say, “I don't have to panic right now because my portfolio fell 20% because my goal probability has only fallen from 98% to 93%, so I'm still in good shape, and I don't have to panic sell,” which is one of the biggest problems a lot of your listeners will face. The moment the market drops, their clients start calling up and say, “Take my money out of the market!” <laughter> They don't have any metric right now to say, “Look, your goals are on track. Don't worry.” In many ways, viewing the portfolio metric as a probability of reaching a goal might actually be a lot more valid.

GE So, it can work in reverse as a client management tool in addition to managing the client's goals?

SD Correct.

GE I don't want to be too greedy with your time, so perhaps we should wrap it up here. Sanjiv. Dan. Thanks for walking us through your work. Anyone who's listening to this should know that we have links to their papers and some other resources in the notes for this interview. I really recommend that you go and check it out. Guys, thank you so much for taking the time.

SD Thanks a lot, Andrew, and good to meet you.

DO Thank you. It was a lot of fun.