The Exchanges

Q And so you could have gone to a hedge fund. You didn't. So what did you do after your PhD?

A So where you go from a PhD is you either take the academic route or you practice, and there was a lovely mentor I had come to know at the Drucker School, and he said, look, there is so much to learn, and practice is really where I would encourage you to go. So I had a choice, and I decided to go to work for Wilshire Associates in California, and dove into portfolio analytics research development Think engineering and design of systems for institutional investors, and the early ones were asset managers, banks, mostly in Europe. So The mandate for me when I joined was to help to build risk models. So there were fixed income models, there were equity models, but there weren't total fund analytic models. So our group was charged with building total fund analytic models and really putting them into the hands of organizations that cared about the multiple asset class products or their own balance sheets, for example. And so early days working within analytics at Wilshire.

AI assessment note: “I decided to go to work for Wilshire Associates in California, and dove into portfolio analytics”

Q So let's turn a little bit to your time at Farmers. And so when you left Wilshire, you're taking over as a CIO. How did you think about what that mission and purpose was and how you structured that portfolio?

A I joined as head of investments. I was recruited by a guy named Peter Teuscher, who was the chief investment officer at the time. The timing was perfect. He said, I'd like you to join Farmers Group. We need to transform the investment process. It was just after the global financial crisis, and there was a lot of work to do. And so imagine, uh, Balance sheet that had grown by acquisition. There was something like almost 50 balance sheets. And, you know, there's a billion dollars in cash. There were state deposits all over the place. There was a lot to optimize, potentially. And like a lot of institutions, proud of how conservative they were in their investments. So, so if you can't really measure where you are in terms of risk, And that means you can't really manage it. What you would tend to do, and a lot of organizations tend to do, is err on the side of less risk. So our mandate was to transform, put risk to work in a way that could grow the surplus, but manage the risk in that surplus. There was a lot of fun work to do there, because it meant basically taking what was one manager who had existed A long time ago and had moved on to an asset management organization and bring in the right kinds of managers. So we had a lot of work to do to transform the balance shades, to bring in the right managers, to put certain mandates to work, but it was also about process. There were sev…

AI assessment note: “our mandate was to transform, put risk to work in a way that could grow”

Q What other factors did you include in that construction?

A So taking a step back, we had Some, what I would call kind of fundamental drivers of what we picked and also robust or statistical drivers for what we picked. And one of the things we wanted to make sure that we did was pick a sufficient set that people could understand. That we could operate with, that would make sense to the managers as well, and would be statistically robust, would explain returns, most returns over time, long term. So those included an equity factor, a duration factor, a credit factor. We had term, which was really representing inflation risk, commodity, and illiquidity. So very, very simple set. Wasn't just enough for us to think about describing the assets in the portfolio. We also wanted to describe the commitments. So if you're thinking about putting surplus risk to work, you have to think about what are the risks inside of the commitments themselves to changes in rates, term, or credit. Now in property and casualty, it's Fairly short duration. It's not like a pension, but we optimize that way. We said, okay, here are the risks, and we're going to match basically the duration and term and credit, and the rest can be put to work in these other factors in a balanced way. So that's how we approached it, and the framework is not necessarily one that would only apply to insurance or pension. It's a general concept, and if you shifted from an asset allocation…

AI assessment note: “included an equity factor, a duration factor, a credit factor. We had term”

Q What teaching from your parents has most stayed with you?

A My father was a scientist. He was a microbiologist, a research scientist, and I would say, fundamentally, he was an explorer. And we grew up in Arizona before a lot of people were there, and he would take us on the weekends, all three of us, and just go out, and we would experience things, and it wasn't necessarily like there was a plan. We just, we enjoyed, and we experienced it, and I remember having a conversation with him once. I don't know, I think it was in graduate school, and maybe this is the time when you start to ask your parents, like, what do you really think about the world? And I said, well, if you could have changed anything about what you did, what would it have been? He said, I would have taken more risk. I would have tried more things. I would have put myself out there. And I took that to heart.

AI assessment note: “I would have taken more risk. I would have tried more things.”

Q What's your favorite piece of reading? Could be favorite book or maybe something online these days.

A So piece of reading right now, there are two things that I'm doing. One is that I'm reading about the history of Rome, Roman civilization, and I'm paying in particular attention to the different leadership through time in Rome and the bigger cultural questions. That they had at the time and the stories that they were telling themselves. It's fascinating to me. That's number one. But number two is I tend to track it. This is where I do train my bot. Well, I tend to track and like to read things about physics. And in particular, I'm interested in entanglement. I don't know if you know what that topic is, but I love reading about it, and black holes, and like, that crazy world of science that's well beyond my reach, but definitely worth exploring. I would if I could.

AI assessment note: “there are two things that I'm doing. One is that I'm reading about the history of Rome”

Q What else was going on in the quantitative world at that time?

A Oh boy. I think, okay, so this was 98. So this is maybe not so early days of hedge funds. Certainly it was the time was a precursor to, as you might imagine, day trading. The internet was just starting to really make itself known to the investment community. And the time was an interesting one of students who had come around the world into this program thinking that, you know, they were going out to the next hedge funds or trading platforms. A lot of them, funny enough, actually took their stipends and day traded and believed that ultimately until 2001 that they knew how to trade markets. So a very interesting time and a lot going on in the field of finance too.

AI assessment note: “this was 98. So this is maybe not so early days of hedge funds.”

Q So when you first dove into that work, what did you see just coming from your own training that you would quickly conclude, well, you know, like, that's stupid. Like, why did they do things that way?

A First and foremost, I was surprised. Really surprised that the mean variance optimization framework was applied directly and simply. Like, even from early days, it didn't make sense to me that these concepts, really, the point was to say something about how risk should be perceived, that it's systematic. These are lessons about how to think about risk, not necessarily to apply it. But I did see those applications everywhere, and I thought it was, it was challenging. And then of course, different and away from that, there were people who didn't have any tools. So I felt like from very early days that there was this massive gap. Between theory and theoretical applications direct from the source to using nothing and maybe around the edges, some tools for attribution, but not in terms of real portfolio construction. I think that was probably early days of those tools coming to market. Think Barra, think, you know, but yeah, it was surprising to me, and so I thought, well, there's a lot to learn here. There's a lot of work to do here, and Getting into that, that work, and there's a kind of a little side story here. At one point, we undertook a strategic relationship with the Bank of New York. It was a really fun project that we worked on, and we involved ING and Harry Markowitz, and we did a survey, a risk survey of a lot of people and a lot of different institutions, Clients of Wil…

AI assessment note: “Really surprised that the mean variance optimization framework was applied directly and simply.”

Q In your initial work with this, have you seen the conversations in the boardroom change from the old, why is that manager underperforming to, well, if we take that manager out, what does it do for our metrics?

A That's exactly the questions that are being asked. And so we're just getting started. I'm sure we're going to have a lot more use cases to talk about, but because the system allows For the question, what if we add the alpha? Doesn't add anything to the outcomes. Do we really contribute to the outcome? So that is a question that easily can be studied. And as I said earlier, net of those common risk factors, is there anything really being produced is ultimately the question that's, that's embedded in the model and being asked. So one of the pieces of outfit that we share are the factors and what the alpha estimates are. It's very easy to begin the conversation. Well, I'm not sure that that's really productive. I'm not sure that this is really our focus. We're, I think that what's exciting to me is that a lot of the people in the organizations that I've worked with start to say, well, why are we having conversations about this or that? Why are we doing that? Why are we spending all of our time talking about individual strategies and individual deals and individual decisions that a manager has made when we have a, a portfolio that's comprised of maybe 10 or 20 managers and We're not at all talking about where we're headed, and it's never been more important to do that.

AI assessment note: “That's exactly the questions that are being asked.”

Q And so what was this model that you built then?

A My research was on the interbank market for foreign exchange. And I was writing about something called price discovery. I was very interested in the dynamics of information and how people make decisions even then in markets. And so what I was looking for was, could we say something about how people formulate prices? What is important? What kind of information is important? Of course, there's permitted information. There's transitory information to signal the noise. Could we make sense of how things worked dynamically in a particular market setting? And so there's a group called Olson and Associates at the time that dumped basically a million quotes on a tape and gave it to the academic community. And out of that was born some very interesting work. And if you've ever heard of the alphabet soup of arch, garch, egarch, igarch, like this is all about time varying variance. And so this was the application that I had studied at the time. So I probably should have made hedge fund out of that. We ended up in that program, really creating one of the first of its kind in financial engineering, which meant bringing together Some of the professors from the math department with the management professors at the Drucker School and thinking about, well, what would it really require to have someone who's trained well enough to engage in portfolio design or trading, et cetera. So that's the sta…

AI assessment note: “My research was on the interbank market for foreign exchange. And I was writing about”

Q Let's break that down a little bit more, because I think even today, there are a lot of institutions that in some way, shape, or form are using Markowitz's mean variance optimization for asset allocation. So why don't we just start with the limitations in Markowitz's model?

A I think it requires us to really take a step back and consider what the investments are actually there to do. So I always start with purpose. Let's think about institutions versus retailer or high net worth, but institutions, there are a lot of different people around the finance committee or investment committee table, if you will. All different backgrounds, different perceptions, experiences, etc. But they tend to agree why they're there. We're here to make grants. We're here to fund a pension. We're here to make sure that claims can be paid. So all, everyone agrees on that. But a lot of the people around the table are not financial experts. They're non-professional investors. And so, when you make the leap from purpose to objectives, it's very hard. Well, exactly what does that mean? How do we carry out the work is what the objectives are. And so, well, that means we're going to pay out X percent per year precisely, and we're going to grow by Y, and we're going to cover inflation. And we also have these commitments that we're making. Well, we also don't want to blow up our balance sheet. So these are considerations that become difficult for non-investment professionals to engage. So what do they do? They turn to experts, consultants, and a consultant will tend to use a model Like a mean variance optimization model. And so you can imagine why. The motivation is actually, it's…

AI assessment note: “ends up really by default saying, you know, these are your two investment objectives.”

Q And what about in like a traditionally thought of venture capital fund? So I don't know, you know, a return stream with like, whatever it is, five bagels, three not so good, and one astronomical return. Like, I don't understand how you can think about mapping that to the same types of factors you'd use in the public markets.

A That is so specific to the fund that you're working with. You really have to hold an index, something like 500 securities in order to know that you're going to actually hit one of those .5 home runs. So anyway, the work with respect to say private equity, bio venture capital, the approach there has to be one by one by one. And so it becomes a conversation about, well, what kind of access do I have? These are more qualitative discussions than they are quantitative discussions. What kind of fund are we a part of? What's their track record? And what are their expectations? Net of fees. What is the chance that I'm really going to have something that's unique here? Could I model this as levered public equities, for example, if I know the leverage of the fund, right? Know the leverage, the leverage inherent in the positions they hold.

AI assessment note: “the approach there has to be one by one by one.”

Q So how did you transition from that one team to whatever it became?

A So we, we said, all right, let's take a look at where we want the risks to be. And this meant, all right, let's get clear about what our objectives are. And then ultimately, when you're talking about risk and risk objectives, you're really talking about the available financial resources. So what can we actually put to work? We're sharing a balance sheet. With the insurance business itself where, and those risks are much larger. So we had to have a place to start. We said, all right, we're going to have several billion that was available in terms of surplus. And then think about where maybe we don't have exposure, for example, to the liquid assets, number one. So is there an illiquidity premium? Is there a risk that we can, we take on here? So thinking about the different common factors that we wanted to put to work.

AI assessment note: “So we had to have a place to start.”

Q to the show. I'm curious how in a portfolio of external managers, once you've identified these factors, there has to be some mapping of the information you have at your disposal. In some cases, it might be security level, and in some cases, it might just be a, an infrequent return stream. How do you bring that together to have conviction that your exposures to a certain factor are accurate?

A That's a great question. When we went through the manager search process, one of the things we asked for was a representative return stream, which is just one observation, time series observation of the distribution of that investment process. Reflective of how they work and how they interact with markets. And so we would gather that information and run it through our model. So we were analyzing managers performance as a part of the decision process. Where were the factors in the strategy? Which factors? Were they significant? Were they not significant? Net of these basic common factors, was there something else that they were producing? Could we call that alpha? We weren't necessarily so interested in, again, defining every single one of the factors, but broad brush, Could we take that into consideration, and were they productive? Net of some common factors, net of cost, was there something left? And there were a lot of managers that actually went through that process and had some performance, and that, to our mind, was, was useful information. It would suggest to us that this is alpha, and that it's consistent through time, and here are the factors. So we never ran those analyses in a vacuum of We always share the results with the managers that were presenting to us. And in many cases, they were surprised. They had never done this analysis in their own portfolios. And what we…

AI assessment note: “we asked for was a representative return stream... and run it through our model”

Q How reliable have you found the numbers and the math that underpins the model that spits this out?

A So let me take a step back and describe how we've approached this. So PI as a framework is the decision framework can be parameterized as a model in a number of different ways. It doesn't have a prescribed distribution or say whether or not you have to use securities or returns, or it's a very general framework. So what we've done is translated The pie decision framework into a system that uses what's called a regime switching model. Our goal was to not necessarily predict what was going to happen over the next five years, but give the range of potential outcomes in a way that would give a path experience that would be realistic. So what we have Built really is a model that accommodates kind of two states of the world. One is kind of a normal market state. The other is a crisis state. And so there's a probability that you'll be in one state and move to the next good state moving to a crisis state. And then when you're in the crisis state, the probability that you move back. So it is possible to estimate the probabilities of being in those states using a very long history. And we know the probabilities change through time, but we started in a particular place, which was to use as much history and accommodate as many crises and policy responses, frankly, that have happened over this time that we have data. And so those two, if you will, regimes and the probabilities are informed …

AI assessment note: “So let me take a step back and describe how we've approached this.”