In this episode I speak with Martin Tarlie, a Portfolio Manager at GMO spearheading their work on Nebo, a goals-based investment platform.

Martin describes Nebo as, “bridging the gap between financial planning and portfolio management,” with a key innovation being the reformulation of risk from volatility to not having what you want/need when you want/need it.  In other words, constraints on both wealth target and horizon.

This reformulation of the core problem introduces a number of complications to the portfolio optimization process.  For example, under classic power utility, lower volatility is always preferred.  But if you’re an investor expecting significant shortfall with respect to your wealth targets, increased volatility may be something very much worth pursuing.

We spend plenty of time in the weeds discussing topics such as: the limitations of dynamic programming via backwards indication, the term structure of return variance, ergodicity economics, and portfolio selection sensitivity to utility function choices.  And while these are all important details, at the end of it all, what Martin stresses most is that it’s the reformulation of the problem being solved that ultimately leads to a more pragmatic solution for allocators.

Please enjoy my conversation with Martin Tarlie.

I'm Corey Hofstein and this is Flirting with Models, the podcast that pulls back the curtain to discover the human factor behind the quantitative strategy.

Corey Hofstein is the co-founder and chief investment officer of new found research due to industry regulations.

He will not discuss any of new found research as funds on this podcast.

All opinions expressed by podcast participants are solely their own opinion and do not reflect the opinion of Newfound Research.

This podcast is for informational purposes only and should not be relied upon as a basis for investment decisions.

Clients of Newfound Research may maintain positions and securities discussed in this podcast.

For more information, visit thinknewfound.com.

In this episode, I speak with Martin Tarley, a portfolio manager at GMO, spearheading their work on Nebo, a goals-based investment platform.

Martin described Nebo as quote, "'Bridging the gap between financial planning and portfolio management, end quote, with a key innovation being the reformulation of risk from volatility to not having what you need slash want when you need slash want it.

In other words, constraints on both wealth target and horizon.

This reformulation of the core problem introduces a number of complications to the portfolio optimization process.

For example, under classic power utility, lower volatility is always preferred.

But if you're an investor expecting significant shortfall with respect to your wealth targets, increased volatility may be something very much worth pursuing.

We spend plenty of time in the weeds discussing topics such as the limitations of dynamic programming via backwards induction, the term structure of return variance, ergodicity economics, and portfolio selection sensitivity to utility function choices.

And while these are all important details, at the end of it all, what Martin stresses most is that it's the reformulation of the problem being solved that ultimately leads to a more pragmatic solution for allocators.

Please enjoy my conversation with Martin Tarly.

This is going to be probably the most differentiated episode of the season, although it's been a pretty odd season, but I do think this is going to be an exciting 1 going from the traditional deep in the weeds of investments to try to cross that chasm between investments and financial planning.

So I know that your current work, you've been deep in the lab working on a project called Nebo, which I'll have you explain in a bit, but I know that your current work is sort of not your first iteration of your career.

So I was hoping we could start with a little background and you could explain how you got to where you are today.

Yeah, So I've sort of done a bit of a random walk through life in a way.

Started out in theoretical physics, went from the sublime to the ridiculous, founded a golf company which actually in some ways is not that dissimilar from Nebo, but then found my way into investing.

I was actually a fundamental analyst for 4 years at an investment fund in the Chicago area.

When I started, I didn't know what a debit and a credit were, but I went to business school and worked with an amazing stock picker who taught me a lot about fundamental investing, but I wanted to combine a lot of what I had been doing on the fundamental side with this sort of quantitative background that I had from my physics career and that's how I ended up actually in Boston at GMO in 2007, where I really sort of started combining concepts from fundamental investing with a lot of quantitative ideas.

And sort of that ultimately does tie back into Nebo.

So let's maybe take a step back before I dive into all my questions.

Can you quickly high-level explain what is NEBO?

So NEBO is a asset management asset allocation portfolio construction platform that solves what we perceive to be a fundamental problem in the industry.

And that's really the gap between financial planning and asset management.

So quote here, you describe it as the gap between financial planning and asset management.

And if it's such a big problem, why hasn't it been solved yet?

So we've been out talking to advisors for now since really 2019.

We didn't know it was a gap when we started.

Just a little bit of background in terms of how Nebo got started.

I've been working on this for the better part of 10 years.

So back in 2013, I was actually in the equity group at GMO, but I'd been doing a lot of asset allocation work and Ben Inker, the head of asset allocation at the time, came down to my office and said, hey, we have a large corporate client who they are migrating from a defined benefit plan to defined contribution and they want to know if we have any thoughts about glide paths, maybe you want to look into glide paths.

So I started looking into glide paths and the first question I asked was, how are the purveyors of glide paths, how are they building them?

And they didn't really explain what they were doing.

So that sort of forced Ben and I to go back to the drawing board and to start thinking about the risk, right?

And so I, you know, later in this podcast, we'll get into what that means.

But basically through this process, we started thinking about risk differently.

And as we started thinking about risk differently, we started engaging with financial advisors because We had tried to apply some of these concepts in the institutional space, but for a variety of different reasons, primarily the retirement focused plans in the institutional space, their primary consideration is they don't want to get sued.

So they're not really early adopters for new ideas.

The other thing that we learned is that these concepts are not really about a particular portfolio per se.

They're much more about the umbrella idea.

And in order to express that idea, you need a platform.

So as we started going around and talking to advisors, they were telling us that there is a gap between what they do on the planning side and what they do on the portfolio side.

Now, there are luminaries in the industry who have also been talking about this problem.

Eric Clark, for example, has talked about the industry has focused so much on the investment problem.

It hasn't focused enough on the investor problem.

Charlie Ellis, If you go back and look at his book on investment policy from 1985 and you read chapter 4, he'll talk about the paradox that is haunting investment management, and that paradox is really around how do you balance short-term considerations with long-term objectives?

How do you know that those long-term objectives are even meaningful?

And that's essentially the problem that we're solving.

But when we go out and we also talk to RIA consultants, so they're a group of professionals where what they do is they work with financial advisors and helping them to understand how to solve their problems better.

When we talk to those people and we ask them about this gap between financial planning and asset management, they say every financial advisor struggles with this issue.

You ask the question, if it's such a big problem, why hasn't it been solved?

So everybody knows it's a problem and they've known for decades it's a problem.

Our view is that the reason it hasn't been solved in part is because academic theory is not suited to the task.

Academic theory starts with the notion that risk is volatility, but if you start with the notion that risk is volatility, you can't really connect the plan to the portfolio.

And we'll sort of get into the details as to why that is.

But the key to unlocking this problem to bridging the gap is to think about risk differently.

Risk is that you don't have the money that you need when you need it.

Now you've described 1 of Nebo's core innovations as redefining risk from being volatility to this idea of not having what you need or want when you need or want it.

So maybe you can answer why is it so critical to redefine risk this way?

How do you think about actually explicitly quantifying that concept?

And what complications does it introduce into the portfolio construction process?

Yeah, so that's a lot of questions there.

So it's not that volatility doesn't matter.

Volatility is an important characteristic of a portfolio.

But when you Think about the risk, for example, that somebody is facing if they're saving and investing for retirement or they're saving and investing for college.

Sort of real-world, meaningful financial problems that individuals or families face.

The risk is that they don't have the money they need when they need it.

And it's just sort of an intuitive definition of risk.

And so going back to when we started this problem in

The glide path purveyors were not explaining how they were building glide paths.

They were essentially, for the most part, using the adage, 110 minus your age, which is sort of an old broker adage.

Now I think we learned subsequently, if you really read in the fine print, Some of them are minimizing the probability that you run out of money, but we can get into the reasons why that can be problematic.

So we just went back to the drawing board and said, let's start with a fresh sheet of paper and ask the question, what is the fundamental risk that somebody saving and investing faces?

It's, well, they don't have the money they want or they need when they want or need it.

It was 1 of the most downloaded papers on advisor perspectives in 2014.

So I think at that time, we sensed that we'd struck a nerve.

And I think part of the reason that we struck the nerve is because the definition of risk that we're using is so intuitive to people.

So really, the reason to define risk this way is to go back to why is it such a big problem?

Well, when you think about whose problem are you solving, when you ask the question, what do you need or when do you need it?

You're placing the asset owner at the center of the problem.

And that really to us is fundamental to the appropriate approach.

If you're going to take the right approach, solve the person's problem who you're focused on, and that's the investor.

It's not the provider of investment products that you're trying to solve their problem.

You're trying to solve the investor's problem, and that gets to Eric Clark's point.

So 1 of them is, well, how do you quantify falling short?

So when you say, what do you need and when do you need it?

Well, what if you fall short of what you need and when you need it.

That actually comes up a lot when we deal with very sophisticated, quantitatively oriented investors.

But we solve that problem actually in a very simple way.

But 1 of the complications is, and 1 of the ways that you have to straddle the line here is, when I've articulated what do you need and when do you need it, I've also said, well, what do you want and when do you want it?

And there can be an important difference between wants and needs.

And you may feel very differently about falling short of a need than you do about falling short of a want.

But when you go to actually practically implement a problem, we've never come across anybody who has been able to quantitatively specify their utility with respect to shortfall versus their utility with respect to gains.

Now, prospect theory has done experiments on this.

We essentially utilize as a default assumption in our platform, something that is very prospect theory like, and it's very simple.

It says that if you fall 20% short of your target, you feel twice as much pain as if you fall 10% short of your target, and you're sort of equally happy with being above your target.

Well, we don't really know, but nobody's ever come back to us and said, I have such a strong view that I want to change that.

The nice thing about that is it straddles the boundary between wants and needs in what we think is a very sort of effective practical way.

Can you talk a little bit about how classic finance theory has solved the single versus multi-period optimization problem and why that this classic approach breaks down when we start to introduce things like wealth targets and investment horizons and asymmetric preferences?

There's sort of a couple of ways to answer that.

I'll start with the single period problem, right?

So although we didn't develop the underlying concepts, starting with a sort of econ

utility-based approach, it was much more Risk is you don't have what you need when you need it.

What's intuitive, this sort of idea that if I'm 20% below, I feel twice as much pain as I intend.

We sort of started with that, but then realized over time that you could actually view everything we did through conventional sort of standard econ theory.

Say okay, what I want to do is I want to introduce the target.

You know, I've got a dollar today, maybe in 5 years I want a dollar 25.

That would be an example of introducing a target.

It turns out that if you start with standard power law utility function, which is by far the most common utility function used in financial economics, nothing interesting happens if all you do is introduce a target.

What you also have to do is you have to introduce asymmetric preferences.

You have to say, I feel differently about being below the target than I do about being above the target.

In sort of standard risk aversion terminology, your risk aversion parameter below target has to be different from your risk aversion parameter above target.

So that's a complication that single period sort of finance theory doesn't really deal with because it's always dealing with sort of the standard power law utility function.

That's the sort of standard mean variance optimization framework.

When you get into the multi-period problems, now you have a whole different host of problems because you have a much more complicated problem that you have to solve.

And even in the sort of traditional utility functions, those problems are really hard to solve.

You know, the multi-period optimization problems are really hard to solve, especially when you have a lot of assets and you have long investment horizons.

So talking about classical power utility functions, 1 of the things about using power utility is that lower volatility is always preferred to a higher volatility solution.

But that's not the case when you introduce wealth targets.

I guess you could say like the intuition here is if you have a wealth target that you're trying to hit and you have very low assets, well, you might need very high volatility.

You might need to go buy lottery tickets to actually have some sense of achieving that wealth target.

Can you talk a little bit about why that doesn't violate common sense, why that is an intuitive answer?

I think the best way to explain that is by an example, but the basic result is that once you introduce asymmetric preferences, once you care about shortfall differently than you care about surplus, you can end up in basically situations where the optimal approach is to seek out volatility.

Volatility can actually be a good, not a bad.

Generally, this happens when your odds of achieving your target are relatively low.

So I'll give you my favorite example, real world example of how this actually played out in the real world.

So Fred Smith is the founder of Federal Express.

And in the early days of Federal Express, he found himself with a real problem.

The problem was he needed to make payroll and he didn't have enough money.

He needed a lot more than he had and when did he need it?

Nobody was gonna lend him money and There was no financial asset that he could invest in where he was gonna get the return that he needed.

He went to Vegas and he played the blackjack tables.

And that's why FedEx is delivering our packages overnight and not somebody else.

Like standard finance theory would say, okay, what happened was Fred Smith went from being risk averse to risk seeking and then back to risk averse again.

What a shortfall-based approach, a what-do-you-need-and-when-do-you-need-it approach, if you want to minimize the risk of falling short of your target, actually what the shortfall optimizing approach is in the circumstances where you need a very high return in a very short period of time, volatility seeking is your best strategy.

It's a much more intuitive, coherent explanation for the behavior that we observe.

1 of the key inputs to many financial planning softwares is the capital market assumptions.

Now, Nebo doesn't impose an investment schema upon the users, but I know GMO certainly offers its own in-house views.

And 1 area you've written about, which is highly relevant to multi-period optimization, is this idea of the term structure of return variance.

I was hoping you could talk a little bit about what the realized term structures of return variance for stocks and bonds have looked like over time and what are the implications of that for multi-period portfolio construction.

Yeah, so if you take equity returns for the US for say the last 100 years or so, and you take the one-year returns and you compute the standard deviation, you'll get something around 17 or

Now you take two-year returns, don't annualize them, calculate the standard deviation, but don't annualize.

Take three-year returns and calculate the standard deviation, etc., etc.

So you do that for all of the horizons, and you plot it out.

And then what you do is you plot out, you take the one-year standard deviation, the 1 year that's 17 or 18% volatility, and you multiply by the square root of that investment horizon.

So the square root of that investment horizon comes if you assumed that stock prices followed a random walk.

What you find empirically is that the measured standard deviations for equities will fall below the random walk line.

So as investment horizon grows, it grows more slowly empirically than what you would Observe if stock prices followed a random walk.

Now do the same thing for the real return on bonds.

And it's important that it's real return.

It actually turns out that bonds are a lot harder to model and understand, and we wrote a white paper kind of explaining all the nuances.

But for the real return on bonds, if you go through the same process, you measure the return, the one-year returns, calculate the standard deviations for a 10-year treasury, it's around 7% or 8%, and you do that for two-year returns, etc., etc., and then you compare that to the random walk, what you'll find is that the volatility profile for bonds rises faster than the random walk.

So What that means is that if your return-generating process does not account for these effects, you will overestimate the long-term volatility of stocks and underestimate the long-term volatility of bonds.

That can have 2 really important implications.

If you're running Monte Carlo simulations that use the random walk hypothesis, and to our understanding, almost every sort of out-of-the-box, random Monte Carlo simulator uses a random walk engine, you will get a distorted view of the outcomes.

For stock-heavy portfolios, you will get variances, variations in future wealth that are wider than what you would if you use something that was more empirically consistent.

Furthermore, if you have a portfolio construction process that is sensitive to long horizons, you will end up with bond-heavy portfolios.

So these are very real, meaningful, real-world implications for using assumptions of a random walk model versus something that is more consistent with history.

Now, I would add that when we wrote the white paper, it was really important to us that we not just rely on the statistical results of actually computing the volatilities.

What we really wanted to do was understand the mechanisms that we're driving, because those mechanisms are actually embedded in the way that we model the return-generating process.

So we went to great lengths to make sure that we understood What are the mechanisms that are driving these volatility profiles?

We're not really comfortable solely relying on the statistical results.

For stocks, it's really about the idea that valuations, P-E ratios in some sense are mean reverting.

So you know if you have a PE of 30 it's likely to come down if you have a PE of 10 it's likely to rise.

For bonds it's actually a different explanation.

For the real return on bonds it has to do with inflation and the fact that inflation is a divisor it's not a numeraire.

So historically for complex multi-period optimization problems it's been solved using this dynamic programming and backwards induction approach.

And you mentioned to me in our pre-call that this approach wasn't actually feasible for the problem that Nebo is trying to solve.

Why would that approach not going to be 1 that could solve the problem?

So the standard way that multi-period optimization is taught in finance and economics is to use the dynamic programming approach where you start at the end of the problem and you work your way backwards.

So usually what you have to do is you start at the end and you say, well, I don't know, for example, what my state variables are going to be 1 period prior.

And those state variables are things like, what are the expected return on assets, what are the wealth levels, etc.

So because you don't know what those variables are going to be, you have to solve for every possible combination of those state variables.

So if you have a meaningful number of assets, say a handful, 5 or 6, about a half a dozen, and if you want to solve problems that are meaningfully long,

years, all of a sudden, the number of possible combinations that you have to iterate over becomes intractable very, very quickly.

So the problem just becomes essentially infeasible to solve.

It's called a dimensionality problem, the curse, right?

It's the curse of the dimensionality in the multi-period dynamic programming approach.

So we actually take a forward looking approach when we solve the problem and I wrote a paper that shows for a very well-known problem that is essentially in the class of problems that we're solving, that the forward-looking approach gives you essentially, well not essentially, gives you in this example exactly the same answer for the here and now portfolio.

So I read that paper where you proposed this feedforward open loop solution I believe is what you called it and I'm not too proud to admit that this 1 stretched me a little bit.

It's been a while since I've done that type of work.

It was difficult to grok so I'm just gonna ask the easy question I should have just asked you all along.

Can you explain how this feedforward open loop solution works and then how it helps solve for that problem where you are introducing wealth targets and asymmetric preferences and deals with this problem, this curse of dimensionality.

So as I said, in a dynamic programming approach, you start at the end and you work your way backwards.

There's another approach, which is to say, I know where I am right now.

What if I invest in a particular way where those investment weights are really my unknowns?

And the rates of return on all of the assets are unknowns too.

But I have a return generating process for those rates of returns.

So what I'm going to do is I'm going to project forward and then I'm going to calculate the expected shortfall, if you will.

And so what I can do is I can eliminate the uncertainty in the return generating process by essentially computing expected shortfalls.

So now I have the basic characteristics of the asset.

And here now it's a term structure of expected returns.

I've automatically built in that volatility profile that we discussed because I've got a return generating process that incorporates that information.

And once I've computed my expected shortfall, I now have an objective function, mathematical objective function, that where the only unknowns are, what are the weights that I'm going to invest in over time?

What are the weights of my portfolio today?

Let's say a year from now, what are the weights in the portfolio?

See all Flirting with Models transcripts on Youtube

Martin Tarlie - Bridging the Gap Between Financial Planning and Portfolio Management (S6E13)