My guest is Roni Israelov, CIO of NDVR. Prior to NDVR, Roni was a principal at AQR Capital Management, where he worked on global risk models, high frequency factors, and lead the development and oversight of options-oriented strategies.
Taking a page from Roni’s career and research, our conversation is far ranging. We discuss topics from global asset risk models to the application of high frequency signals to tail risk hedging. While there are insights to glean in each of these topics, I think the conversation helps paint an insightful picture about how Roni thinks about research in general.
Towards the end of the conversation we talk about the new research Roni is tackling at NDVR, a financial advisory firm for high net worth individuals. The role brings new challenges to consider, such as liability management and risk tolerance within the framework of portfolio optimization. Even though the topics differ, I think you’ll hear a very common thread in how the research is performed.
Please enjoy my conversation with Roni Israelov.
Corey Hoffstein 00:00
Okay, you’re ready to go.
Roni Israelov 00:01
Let’s do this.
Corey Hoffstein 00:02
All right 321 Let’s jam. Hello and welcome everyone. I’m Corey Hoffstein. And this is flirting with models, the podcast that pulls back the curtain to discover the human factor behind the quantitative strategy.
Corey Hoffstein Is the co founder and chief investment officer of new found research due to industry regulations, he will not discuss any of new found researches funds on this podcast all opinions expressed by podcast participants are solely their own opinion and do not reflect the opinion of new found 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 in securities discussed in this podcast for more information is it think newfound.com.
Corey Hoffstein 00:53
My guest is Roni Israel of CIO of endeavor. Prior to endeavor Roni was a principal at AQR Capital Management, where he worked on global risk models, high frequency factors, and led the development and oversight of options oriented strategies. Taking a page from Rooney’s career in research, our conversation is far ranging. We discuss topics from global asset risk models to the application of high frequency signals to Dale verse hedging. While there are insights to glean in each of these topics, I think the conversation helps paint an insightful picture about how Roni thinks about research in general. Towards the end of the conversation, we talk about the new research, Roni is tackling at Endeavor, a financial advisory firm for high net worth individuals. The role brings new challenges to consider such as liability management and risk tolerance within the framework of portfolio optimization. Even though the topics differ, I think you’ll hear a very common thread in how the research is performed. Please enjoy my conversation with Ronnie Israel. Tony Israel of thank you for joining me, this has been a podcast a little bit of time in the works, I guess you and I recently collaborated, we’ve been going back and forth as colleagues for a couple years now. I am excited. It should have happened seasons ago. But I’m delighted to have you on now. Thank you for joining me.
Roni Israelov 02:18
Yeah, thanks for having me. I’m delighted to be here. I’ve been listening to your podcast for a while. And it’s fun to be able to join and have this conversation with you.
Corey Hoffstein 02:25
Well, I really appreciate you listening. Let’s start where we always start, which is background for those listeners who maybe don’t know you. Can you fill the people in on where you’ve been and where you are now.
Roni Israelov 02:37
Yeah. So my career and quant started about 16 years ago, I finished my PhD at Carnegie Mellon in early 2017, around April, so right around 16 years ago. And my thesis had three papers, one on liquidity, one on liquidity risk, and a third paper on theoretical econometrics. So that was a pretty weighty paper. And I had an interest in joining the buy side and being active in trading and portfolio construction. And I was pretty excited to get an offer to join Lehman Brothers. And my offer was to join an equity stat, our prop trading desk, which was applying machine learning methods, maybe ahead of its time in their prop trading of equity strategies. So I was excited to get the offer. And two weeks before I was supposed to join, I got a phone call. And it turns out that team was leaving Lehman Brothers. And my offer was transferred to a different team, which was a sell side team instead of a buy side team reporting to Matthew Rothman and quant equity strategies. So it was a pretty big change versus what I was expecting when I was joining Lehman Brothers. But the good news was that Matt Rothman had a buyside background, he had worked in a number of different quant roles. And he was really bringing a buy side framework to Lehman Brothers and running the team. Pretty much as though we’re a buy side team. We were building models in the same way that other quant equity managers would build models and going through all the same type of research, the only difference being we weren’t actually trading the positions. So I joined his team in early 2017. I was there for about a year and a half. And as you probably know, it was a tough time for Lehman at some of the other banks. I still had this goal of being on the buy side and an opportunity presented itself for me to leave Lehman Brothers and join AQR Capital Management. So I was pretty excited about that opportunity. And I left Lehman and joined AQR on September 8 2018. A week later, Lehman collapsed. So it felt like a pretty good decision to make that move. And I spent 11 and a half years at AQR and it was a great experience. I learned a lot. I think I contributed a lot to the organization. session, I had the opportunity to work on many different strategies and many different research assignments, which was really intellectually stimulating for myself. And I think we’ll talk about some of that today. But just to give like a quick highlight, I joined the global asset allocation team. So when I was at Lehman, I was focused on stock selection strategies. When I joined AQR, I moved to global asset allocation strategies with a focus and initial focus on equity country selection. So these were strategies that were primarily run in their hedge funds, alpha seeking strategies. And within that role, I worked on equity factors. But it also provided the opportunity to begin working on risk models that would be applied to the equity country selection models, but then more broadly applied throughout the global asset allocation ecosystem. So that was a pretty sizable project, given the scope of its use within the organization. Having worked on equity factors, most of the factors I focus my attention on were longer term in nature. But at some point, there was some attention being paid to shorter term factors and an opportunity presented itself to help build out a future set of strategies. So I worked on that for a couple of years. Very different framework than you would apply to the longer term factor based portfolios. It was a lot of fun, I enjoyed it immensely. The research as I said, was quite different. But then another opportunity presented itself to join and oversee the options team at AQR and at the time, the options or volatility strategies that AQR was managing are primarily limited to their hedge funds. So they tended to be either tactical or relative value in nature. And that was really the goal to continue to manage and enhance those strategies. But as I was diving into it, I saw an opportunity to try to build a new business for AQR on the volatility risk premium harvesting side. So conducted a lot of research started building a team and supportive that business publishing papers speaking at conferences and developing strategies and portfolio construction methods that we thought were appropriate for those who were seeking to harvest the volatility risk premium. And that basically took me to the end of my time at AQR I left in 2020 February 2020. However, the last year and a half that I was at AQR, I was also offered the chance to oversee the portfolio implementation research team, basically portfolio construction for global stock selection. So in a way I came full circle or left Lehman Brothers working on stock selection, worked on a number of different things that had little to do with stock selection. And then in my last year, year and a half at AQR, I had the opportunity to again work on stock selection. In February 2020. I left AQR and joined endeavour, spelled mdvr. It’s devalued because we’re a modern wealth optimization firm. And endeavor is a firm that serves affluent and high net worth investors. So we are a Direct to Consumer Direct to client advisory firm. I oversee the portfolio management team, the research team, the financial planning analysis team, and I also assist with executive functions. I’m the Chief Investment Officer and the president at Endeavor.
Corey Hoffstein 08:26
Well, I can’t help but notice a somewhat obsessed with timing luck of leaving Lehman, the week before Lehman went under and then leaving your role as head of options at AQR right before maybe one of the biggest VRP blow ups in March 2020. So I would appreciate a head nod next time you make a major career change, I think all of us would like that signal.
Roni Israelov 08:48
Yeah, that was pretty funny. And especially after the fact interestingly enough, there are a couple of interesting components to leaving AQR just before the vault blow up. So my understanding having spoken to some people outside of AQR is they didn’t understand that I left before the ball blow up, and they had assumed I left because of the fall blow up. Not true, you know, I left before of all blow up. So that’s point 1.2 Is we were very careful about portfolio construction involves strategies. And one of the considerations that we had when we constructed portfolios is understanding the tail risk of individual positions having a tail risk budget sizing positions according to that tail risk budget. And also because fall strategies at AQR really did start from the hedge fund framework and, you know, had this notion of tactical views and Relative Value views. You know, we were incorporating some of those views into our portfolios. So when you include a portfolio construction, towards tail risk management and some tactical positioning, I think AQR actually came out relatively unscathed. And that vol blow up especially relative to some of the Are there more news worthy organizations?
Corey Hoffstein 10:03
Well, I’m excited to dive into all this. There’s a lot of heavy stuff to unpack. You’ve had a wonderful breadth of your career so far. And I’m sure that’ll continue. Last sort of high level light question. I’m curious, do you have any great Lehmann swag in the closet that you’ve kept onto a hat? You know, vest?
Roni Israelov 10:18
I wish I have nothing. I’m not good at keeping memorabilia. Unfortunately,
Corey Hoffstein 10:24
I would have been fun stuff to hold on to. All right. Well, let’s start off the deeper questions. With your time at Lehman, you started mid 2007. And then immediately, August 2007, there’s the great quant quake. And you ended up spending a decent amount of time doing some research into this idea of the correlation among different quant equity managers, their convergence and sort of risk factors around that concept. First, can you at this point, it’s been so long, I don’t want to assume all of our listeners know what the quad quake is. So maybe you can provide a little bit of a history lesson as to what the quad quake was. And then can you discuss some of your findings from that project?
Roni Israelov 11:05
Yeah, and it was an interesting way to start my quant career, I joined Lehman Brothers. And a few months later, you had this huge event. So basically the quant quake, that’s a hard expression to say, or the quant meltdown was an unprecedented widespread, statistically improbable set of losses that occurred pretty much across the long short equity quant community. And I think that is what was concerning the magnitude of loss that occurred as well as the fact that it pretty much hit everyone. And it was widespread. There was an initial preview and late July 2007, you started to see some stress and small cap value strategies. For some managers drawdowns were starting to accumulate, there was some increase in volatility. We were already writing notes about that there was a lot of curiosity about that at the time. But then the week of August 6, the hammer came down. And basically, you had four days back to back of huge minus five minus six SGD event losses, each day in and of itself was newsworthy. The accumulation of those returns over four days in a row was eye popping and shocking, and really caused a lot of stress in the system. The prevailing theory, I don’t know if I ever heard you know, I don’t know if anybody came up with the official story of what happened. But I think the prevailing theory is that this events started at a multi Strad that saw losses in their fixed income book, I think I’ve heard some argue that it originated in their mortgage book. And that caused them to de risk there are liquid positions. And this is kind of a funny thing. But sometimes, it’s not the area of loss that gets liquidated. It’s the liquid part of a book that gets liquidated when there are losses. And I think that’s what occurred here. So they ended up liquidating, or de leveraging their equity Long, short book. And that led to a feedback loop across quants who are similarly positioned. And there’s an important point in that. And for those who are not necessarily familiar with equity, long, short strategies, the rebalancing behavior for them is very different than an equity long only portfolio. So if I’m an equity long only Portfolio Manager, and I see a loss of 10, or 20%, that loss doesn’t necessarily require a rebalance because my nav has gone down. But my exposure has gone down by an equal amount they track each other. And a rebalance isn’t necessarily required. But if I’m an equity Long, short Portfolio Manager, it’s very different. If my nav goes down by 10, or 20%, because I’m realizing some losses. Well, my gross notional exposure on the long part of my portfolio in the short part of my portfolio is not really unchanged by that. So unless I rebalance the portfolio, and D leverage the portfolio, my risk allocation has gone up as a result of having a lower NAV. Beyond that, depending on how quickly I’m updating my risk estimate of the factors if I just saw five STD events, my view of the risk of the factors has also gone up. So I have a lower NAV, I have an increased view on the risk of my portfolio. If my entire goal is just to keep my risk target unchanged, I have to reduce the size of my book considerably. But some people may want to go beyond that, right because they may have risk controller drawdown control whereby after they start to realize significant losses, they actually want to de risk their book. So you have these force rebalancing trades that can cause a feedback loop especially if a number of quants are similarly positioned. And I think that is what happened over the four day period in August which caused a lot of stress in the market now, August tend you saw a rebound in some of the factors. The problem is, if people had already traded down their book over the four days, because of the losses that occurred, you aren’t able to fully capture that rebound. So you can go through the entire five day period and still end up with material losses after some of the factors started rebounding. And I think you saw incredible dispersion actually across quant managers, depending on the actions they decided to take as a response to the losses that they were starting to realize. So it was an eye opening event for the quant community and a learning experience. And to your question about the research on correlation will naturally lead to a lot of questions about crowding in the marketplace. You know, the fact that a number of quants were seeing similar losses develop and it was causing a feedback loop caused a lot of concerns amongst quants to maybe even you know some concerns among non quants about the volatility that quants might be bringing to the market. So it was actually a pretty interesting place to be at Lehman Brothers because Matt Rothman was very well connected amongst the different quant asset managers. And remarkably, he was able to convince 10 large quant funds to give him their raw underlying positions for their s&p 500 book. And I think also for the Russell 2000 book. In return, he would provide our team would provide some information about how correlated their positions were to their peers. And you can imagine like how stressed these managers must have been to provide this information because this is information you would normally hold very close to the vest and everything was anonymized. So you know, to try to preserve and keep information confidential. But there was enough curiosity about correlation amongst the quads to provide this information. So we ran the analysis, we had quintile ranks for s&p 500 portfolios across 10 different managers, so 100 names in each quintile. And if managers are uncorrelated to each other, you’d expect about 20 names of overlap within quintiles but of course, we didn’t see that what we saw was roughly on average 40 names of overlap from, you know, the long quintile of one manager to the long quintile of another manager, about 40 names of overlap in the short quintile of a manager versus another manager, and about 10 names of overlap, if you’re comparing the long the best quintile of one manager against the worst quintile of second managers. So definitely seeing some evidence of correlation amongst positions, I think it’s hard to interpret is that a lot of correlation or a little bit of correlation, as we were presenting the results to people, I think the interpretations were pretty mixed. Some people were kind of astonished at how similar these positions were others were astonished at how low this similarity was. So I think that’s really open to interpretation. The last finding that I would note that came out of the research, is we were curious about how portfolio construction may play a role in this. So we took those underlying quintile ratings provided by managers fed them through a portfolio optimization engine with a common risk model and typical constraints, one might apply, and looked at kind of the post optimization, position correlation. And we did see that that increased so portfolio construction can also play a role and creating more correlated underlying positions. It’s not just about the underlying Alpha factors.
Corey Hoffstein 18:33
My conspiracy theory has long been totally unfounded. That’s why I call it conspiracy theory is that at that point, all the managers had adopted the same risk factor modeling software, something like bhara. And to your point, it was the convergence on the same modeling software and optimization software that led to the convergence of positions. So totally unfounded. I’d love if someone could prove that but that’s my that’s my crazy tinfoil hat conspiracy theory. I want to stay on this idea of risk. You’ve sort of touched upon it a couple of times, obviously, a key concept. This was something you focused on heavily in your early days at AQR, you actually worked on global asset risk modeling. But your work there found its way across a lot of the hedge fund strategies trend following risk parity, long short, some of the other hedge fund strategies. I’m curious, do you think that risk is sort of something that should be measured consistently across strategies? Or is it something unique to each strategy? And in other words, is there sort of a central truth to measuring risk?
Roni Israelov 19:38
So I think it’s important to start with the underlying objective when building a risk model, what is the purpose of that risk model? I think if the purpose of the risk model is similar, and it’s within a similar strategy ecosystem, then I think there is a consistency of modeling approach that applies but there are very different use cases of risk modeling and I talked to About the fact that I was involved in helping to build or update the risk models applied to global asset allocation. I also worked on risk models and the option strategies, because if you’re going to apply tactical positions and options, one way to do that is to have a forecast of volatility and compare that to implied volatility. In the last year of my time, when I was focused on portfolio Implementation Research at AQR, I was working on factor based risk models within the stock selection portfolio. So I actually ended up touching risk models across a number of domains. And what I would say is, again, the objective matters. So you know, if I think about the tactical option, risk model, in that case, I think it’s important to have a very precise estimate of risk and ideally, measure of risk that relates to the option maturity. So if I have a five day option that I’m considering I might have a different risk model or calibrated risk model for that option versus a 20 day maturity option. If I’m applying risk models to factors in stock selection, then there’s a framework that you would bring in in terms of which risk factors are you including and how are you applying those but your question was about global asset allocation? So let me let me go to that. I think, and I would just restate what I said at the outset that the objective of the risk model is important. So in global asset allocation, one of the primary objectives of the risk model is to help size positions because it’s pretty common amongst quants that the allocation to individual positions is a risk allocation. So a model might say allocate 2% of risk to s&p 500, allocate 3% of volatility to Treasury futures. And in order to provide that allocation, you need an estimate of risk, you know, it relies on volatility model. So given that, what are the objectives of the risk model? Well, one objective is that you are able to realize volatility that is consistent with that target. If I’m targeting 2% volatility, I would like to realize 2% volatility, and I would like to do so consistently. But that’s not the only objective. I mean, if I’m trying to satisfy that objective, it might make sense to have a quickly updating a short term risk model. But there are competing objectives. If I’m going to target risk in equities or fixed income, I will have to buy and sell the underlying instrument as my view of volatility changes. So there’s a turnover cost to risk. And if I have a rapidly updating risk estimate, I might have to realize a lot of turnover. And that may be impractical, there may be a tail risk or kurtosis impact because if I have a short term risk estimate, by the nature of it being short term, it has estimation error, and there’s a chance that I underestimate risk, in which case I have more leverage or a larger exposure than you might want to otherwise. And that can introduce tail risks. So I think when you’re estimating risk, there are competing objectives. And for that reason, it can make sense to build efficient frontiers across this objectives. And one thing that I found is it makes sense to have a consistent framework across a set of strategies if they have the same underlying goals, in this case, you know, within global asset allocation. But the other thing that I found is, when you have that consistent framework, you actually find a pretty robust pattern, that the efficient frontiers across these objectives look pretty similar across assets. And they also end up looking pretty similar across asset classes, which means that it’s possible to somewhat simplify the overall estimation and view the calibration exercise is a large panel, rather than as a set of individual calibrations for each underlying asset or asset class. So I don’t know if I would necessarily say there’s like a central truth. But there’s a robustness that you see when estimating or calibrating the service models, which allows for some structure and simplification, and common application across assets and asset classes.
Corey Hoffstein 24:03
I think about it from the utility function of the asset manager running a number of strategies, knowing that there’s going to be estimation error, the risk of bias and that estimation error propagating among all the strategies, it almost seems like nefariously in a certain way, you might want to introduce noise randomly to your risk measures among different strategies just to try to reduce the correlation of estimation risk among the different strategies. You don’t want them to all be wrong in the same direction. In other words,
Roni Israelov 24:32
right, I hear what you’re saying. I mean, a lot of times when I think about modeling and estimation, what I’m really concerned about is the question of like, what if I’m wrong, like, what if the estimate is noisy or the model is wrong? How robust is the portfolio construction to those errors? And if it’s not very robust, I think that’s a concern. And if there are ways to increase the robustness, I think that’s attractive.
Corey Hoffstein 24:55
Alright, we’re gonna jump around a bit here. You got a lot of papers I want to touch on. You’ve done a ton of broad Research. So let’s dive into some of your work on higher frequency factors. So this was an area again that you focused on at AQR, some of those higher frequency factors multi day to intraday, I want to get your opinion on how alphas at this horizon differ from the alphas we typically think about it slower time horizons, are these orthogonal signals to sort of those slower style premia, or are they just some of the common premium applied at higher frequency.
Roni Israelov 25:30
So I think in some instances, they can be similar. So there are a number of the style premia that are primarily price based call it momentum, one frequency or potentially reversal at another frequency. And you can imagine applying similar price base signals at a higher frequency, you know, short term momentum, or short term reversal. And structurally and thematically, those are going to be pretty similar to the longer frequency signals. Just because they’re similar doesn’t necessarily mean they’re correlated, though, if you have a factor that’s turning over annually, or semi annually, and then you have another factor that’s turning over daily or weekly, almost by construction, on average, those are going to be very low correlation factors to each other, which makes them pretty diversifying. But I think when you have longer term factors, oftentimes, you’ll incorporate some sort of fundamental information value or otherwise. And I think those oftentimes don’t work quite as well, in the higher frequency domain. And when you go very high frequency intraday, let’s say, and everybody’s definition of high frequency is different, you know, so we’re not talking latency arbitrage here, but intraday factors. I think, by and large, the focus on those are trying to predict and take advantage of structural flows that are occurring in the market, you know, if there’s some sort of rebalancing activity that’s mechanical, that leads to certain flows, and that’s for castable, then I think those are pretty common, higher frequency factors that one might seek out, and this type of strategy, which are pretty different than the style factors that would be applied and larger timeframes.
Corey Hoffstein 27:12
From your work, you ended up co authoring a paper with Michael Katz called to trade or not to trade, in forex trading with short term signals for long term investors. Can you talk about the problem you were trying to solve in this paper and the proposed solution?
Roni Israelov 27:27
Yeah, so the underlying motivation is that sometimes when you’re doing research and long term factors, a byproduct of that research is short term factors wasn’t the goal, but they come up in certain ways. And as you continue to work on the factor research, you start to develop maybe a collection of the short term factors, and they may be statistically significant, there might be an argument in favor of them, but their performance is not quite strong enough to survive trading costs. So you might see a short term factor with perhaps a gross of trading cost Sharpe ratio of one or a half or whatever it might be. And then as soon as you apply trading costs, you’re at a minus one or minus two Sharpe ratio. And that can be pretty frustrating. It feels like it’s inappropriate to throw away information when building portfolios. So that had happened and that was on our minds. And we were seeking out a way to incorporate that information in portfolios. And the idea was to use the short term signal to only affect trades, and to use the long term signal to determine the portfolio itself. So as an example, you can imagine a long term signal would tell us that we want to be long, s&p 500. But then a short term signal might tell us that the expected return or alpha of s&p 500 over the next week is negative. So what we would do is we would say, well, let’s allow that short term signal to tell us not to buy s&p 500 yet, and in a few days or a week, We’ll reevaluate and we’ll look at the short term signal. And if it’s negative, we’ll wait another week. If it’s positive, then we can go ahead and put on the trade. And we call that informed trading. I’m not sure it was like an entirely novel idea. I mean, if you read some papers, sometimes people allude to this idea of applying short term signals in such a way, but to our knowledge, we had not seen yet any analysis that had formally investigated this to see how efficacious it was in portfolio construction. So we wanted to test this not for the purpose of publishing a paper that itself was a byproduct of the research. But you know, we were doing this because we were interested in the application of it. And as I said, I was at the time focused on global asset allocation in equity markets, equity index markets, so that’s where we tested it, we developed a framework to test it in developed equity markets. And then we also created a simulation environment where we could test this and more of laboratory setting to better understand his properties with a similar type of cross section, as you might see and developed equity markets. And I think the findings were actually pretty fascinating. So if you typically think about adding a factor to a model, so you have an existing model with maybe 10 factors in it, and I have a new factor that I want to add to the model, what happens is, is for me, to add weight to the new factor, I have to reduce exposure to one or all of the existing exposures. So it’s a trade off, I want to add some information, I reduce exposure to the existing factors, I add exposure to the new factor. And that’s, you know, pretty standard practice. But what we found with informed trading, is that because the short term signal is effectively reducing turnover, kind of on its own, because you’re just canceling trades. And sometimes by the time the short term signal says to buy, your long term view is changed, so you don’t want to buy it anymore. So the application of informed trading reduces the turnover of the strategy, it allows you to be more aggressive in trading than you would have otherwise, because of that feature. And also, because you’re now adding some short term information to the model. So what we found is, if you optimize the model without using informed trading, and then add this informed trading algorithm and re optimize the model and recalibrate how aggressive you want to be, in terms of your trading, you end up being a little bit more aggressive on trade, and your total trading costs tick up a little bit. But you buy a lot for that additional turnover. And you actually, and this is what was remarkable, you actually increase your exposure to your long term signal while adding exposure to the short term signal, which really changes things, you know, typically, you have to take a haircut on your exposure to your existing model to add the new factor. And here, you are able to increase exposure to one while adding exposure to the other. And we found that to be pretty exciting.
Corey Hoffstein 32:01
definitely exciting, and fascinating that you shared it with the public at large. Right, a consistent theme in your career is that you’ve written a large number of research papers that have found their way into leading industry journals. I think many of us researchers are greatly appreciative of that. But my question for you is, and perhaps it’s an impossible question is, which is your favorite paper and why?
Roni Israelov 32:26
Yeah, so I think my favorite paper, what has to be one that I wrote about three or four years ago called give credit where credit is due What explains corporate bond returns, and let me motivate what drove the paper. So at the time, I was overseeing options, strategies, and as I mentioned, volatility risk premium harvesting was one of the strategies that we were trying to build a business around. And when we call around and talk to asset allocators, and you know, people in the industry, one thing that I noticed is that it seemed to me like there was incredible comfort around allocating to corporate bonds and investment grade and high yield corporate bonds, but particularly investment grade bonds. I mean, it was almost a default allocation, and people wouldn’t give it a second thought. At the same time. There is incredible discomfort in allocating to a volatility risk premium harvesting strategy, because picking up pennies in front of a steamroller, the tail risk, I mean, you can imagine all the arguments. And I was fascinated by that, because it appeared to me to be a contradiction, a corporate bond, a short volatility, it has very similar exposure. And I couldn’t understand why there was such comfort, with shorting volatility within a corporate bond and such discomfort with shorting volatility and option markets. And I thought it would be interesting to write a performance attribution paper on corporate bonds to help better illustrate the idea. So that was my motivation for the paper. I’d say the way the paper motivates it is by an eye towards the Merton model. And the Merton model says that there is an economic equivalence between a corporate bond and a government bond with a short put option on the firm. That short put option on the firm introduces exposure to the firm. And you see that of course, corporate bonds have positive correlation to equities, but it also introduces a short volatility exposure. So that is the underlying Merton model, a number of papers have looked into the Merton model. And the typical approach that they take when they do this is they will say that there is a relationship between corporate bonds and volatility markets. And usually, the way they’ll kind of attribute this relationship is by mapping credit spreads to implied volatility, or mapping implied volatility to credit spreads. What I had not seen yet is a paper that looked at this from the return point of view, and you know, what relationship do you see in return? And that is what I wanted to explore. So I looked at a large cross section of corporate bond indices across the credit dimension, so investment grade corporate bonds and high yield corporate bonds, and across the maturity dimension, so, short term, intermediate term and long term, so a large collection of these corporate bond indices, and I was pretty careful about a performance attribution. So, I wanted to understand how much performance were these bonds getting from their equity portfolio allocation. But the companies and these corporate bond indices aren’t the same, necessarily, as the companies in the s&p 500, you have different types of characteristics. So, I built a constituent waited equity portfolio to align the characteristics of the equity portfolio that I’m using as an explanatory variable with the corporations that you see, and the corporate bond indices. And I did the same thing with a single name options. And as I said, you know, I’m pretty careful performance attribution. And when you go through all of that, the first finding, which is attractive is that all the exposures that come out of that just make intuitive sense. They match your intuition you see exposure to equities, you see exposure to duration match bonds, you see, short option volatility exposure, you see short bond, volatility exposure, which makes sense because a number of these corporate bonds have callable bond features. And you see that it is those bonds that have callable bond futures that have the short bond vol exposure and those that do not, do not have short bond vol exposure. So it’s very nice to see intuitively, the relationships that you would expect. But the other thing that I think comes out of this is that corporate bonds are very complex. I mean, they, you buy a corporate bond, you think you’re buying like a single instrument, but you’re really buying an entire collection of factor exposures. When you buy a corporate bond, the firm’s tend to be value stocks, if they’re high yield firms, they tend to be high beta, which means you’re loading negatively on Bab, they tend to be junk, which means you’re loading negatively on quality. And if you just do a very careful attribution of this, what I found is that as a whole corporate bonds have exposure to positively compensated risk premium, they have exposure to negatively compensated risk premium, and they have a healthy amount of uncompensated risk as well. And then the question I have is, well, if there are other ways of getting these exposures, why not just allocate to the underlying risk premia themselves and avoid shorting positively compensated risk premium within a corporate bonds. So I construct this idea of a synthetic corporate bond, which essentially aggregates the desirable risk premia and show that when you do that, you end up with higher returns because you’re not loading on negative risk premia factors. And you end up with lower volatility because you’re not taking uncompensated risk or allocating negatively to compensated risk. So lower volatility, higher return higher Sharpe ratio. And I like the paper. And my guess
Corey Hoffstein 37:57
is every pension everywhere continues to allocate corporate bonds.
Roni Israelov 38:01
It’s a very hard thing to sell. Because I think the allocation structure of pensions is not really you know, I mean, the way things are bucketed. It’s not really designed to substitute volatility risk premium for a corporate bond,
Corey Hoffstein 38:16
they’re not exactly necessarily easily fitting into the LDI framework. While sticking with talking about options and volatility. I want to talk about one of the papers that you’re most famous or perhaps infamous for, depending on what circle we’re talking about, which is your paper pathetic protection, the elusive benefits of protective puts. By the way, I have to say, as a side note, all the papers coming out of AQR have the best titles. I don’t know if you’ve got a committee there that helps come up with titles or it’s just part of the culture, but they’re always great. More recently, you wrote another paper called equity tail protection strategies before, during and after COVID, which I thought provided a really interesting decomposition of these different tail hedging strategies. Both papers, and I don’t want to paint with too broad a brush here. So correct me if you think differently, but I think they sort of came to the same conclusion, which is, tail protection is complex and fraught with risk. And I don’t think that’s like a controversial position. It’s a difficult thing to do correctly. Curious what you thought some of the most interesting findings were from this research and ultimately afterwards, what is your stance on tail risk hedging?
Roni Israelov 39:27
That’s a great question. And it’s interesting that you say that you don’t think it’s controversial. I do feel like whenever I re highlight these papers, it does seem like there’s a lot of disagreement about my conclusions. And funny enough, when I wrote the title for this paper, pathetic protection, I did not think the title would survive, and I was pleasantly surprised that I was able to keep that title through publication. To me the motivation for the initial paper was that oftentimes when people think critically about tail protection strategies, they tend to Really focus on the cost side, you know that there is a negative carry? And yes, the benefits you get from child protection are very desirable. But are they worth paying the negative carry? I think that’s the common framing around child protection strategies. And you know, it’s related to the idea that there’s volatility risk premium, I was curious about challenging the benefits side of the equation, I felt like we were too quick, as an industry to assume that there were strong benefits to buying tail protection. And I was not convinced because of path dependence concerns. And this is very similar to your concerns about rebalance timing lock. But options have a maturity schedule. And when you think about buying an option to protect a portfolio, it can work very well mechanically, if the option horizon is aligned with a protection period somebody cares about. So if I want to protect my equities portfolio returns over the month of May, and I buy an option that expires at the end of May, that should do quite a good job. And it’s mechanical, that should do a very good job of protecting the portfolio. Now there is a question about the cost of that. But on the benefit side, I think there should be a lot of comfort, where I think it potentially falls apart is, if I don’t have such a well defined period of protection, if I’m thinking about tail protection as an evergreen strategy, you know, and I’m a 20, or 30 year investor, and my concern is peak to trough draw downs. It is not clear to me that buying systematically options for protection actually helps to reduce peak to trough draw downs. And that is what I wanted to test in the paper. So I did it two ways, this is going to sound like a theme. So I did it with data and using s&p 500 options and, you know, evaluated the strategies and looked at the performance of the protection in terms of its ability to improve peak to trough drawdowns. And I found that when you buy protection, and you have longer term horizons that the peaks and troughs drawdowns of the protected portfolio are worse than an unprotected portfolio when both are sides to have the same comparable return. And that was largely driven by the path dependence issue that I talked about, but also driven by the fact that there’s a volatility risk premium and that protection is expensive. I wanted to isolate the benefit side and get rid of the expensiveness idea. So I set up a simulation environment where I simulated equity returns geometric Brownian motion, simulated option prices, but I simulated the option prices where there’s no volatility risk premium where there’s no alpha. So I was concerned about the benefit side of the equation. And I wanted to test whether protection strategies can be beneficial if there’s no volatility risk premium in the options market. So I set up the simulation environment, I simulated equity returns as a geometric Brownian motion simulated option prices with no volatility risk premium, so there’s no negative alpha when you buy them. And then I evaluated the efficacy of the protection strategy under that framework. And the findings were generally pretty similar that the peak to trough drawdowns of the protected portfolio over a longer horizons were worse than a peak to trough drawdown of an unprotected portfolio. So I thought in its own right, that was a pretty interesting paper. It may be controversial, you know, I gave it an aggressive title to hopefully generate some interest. And then lo and behold, the next major events that happened in equity markets was the COVID crash of 2020. And the trough of equities during that COVID. A crash basically occurred on the option expiration date, and put protection strategies nailed it. They did an incredible job of protecting equity portfolios against that crash. And you know, this is Murphy’s Law in action. It is what it is. I mean, my claim was never that protection strategies are universally bad. It’s more of a distributional argument. Sometimes they work sometimes they don’t work, but you know, it’s not reliable. And on average, they might make things worse, a couple of years past and then we saw a very different kind of drawdown in 2022, where it was just a slow and steady grind down not really any material or notable spike in volatility. But the losses in 2022 were notable. But what was more notable was that it seemed like no protection strategy really helped in any way, shape, or form. If you were buying put options for protection, it just made things worse, like they were rarely expiring in the money. The options were still priced somewhat expensively, so the returns of the protected portfolio were worse than the returns of the unprotected portfolio. If you bought vix futures, there was no notable spike in vix futures so that didn’t really help So I thought that was interesting. And that’s what led to this retrospective paper, which was about equity tail protection before, during and after COVID, which was intended to really provide context and expand a little bit the aperture because the pathetic protection paper was focused only on put options. And I thought it would be interesting to talk about the protective features of a number of different commonly used option strategies for tail risk hedging. One thing that I think that was on my mind, as I was writing that paper, is whenever I talk about the difficulties of some options to protect a portfolio, it feels like the response that I always get is, nobody would protect their portfolio in that way, we have a much more advanced protection strategy. And what I wanted to do is look at a bigger collection now until you can literally evaluate an infinite number of combinations, I think you can never satisfy that criticism. But it was interesting to me that when you look at now three or four different strategies, you still see kind of the variability and outcomes. And to me, I think that just highlights the challenge that even if it’s true that there is some protection strategy that is very effective, it requires people to be able to identify it and find it. And it’s not necessarily obviously the case that people can do that very well.
Corey Hoffstein 46:23
I want to flip to the opposite side of the coin here talk a little bit about shorting ball, which is, you know, again, sort of towards the end of your career at AQR, you were leading that team, you helped build and oversee the option oriented strategies. Curious, from your perspective, how does something like the volatility risk premium or the variance risk premium really differ from underlying asset risk premia like an equity risk premium or a bond risk premium, both from an economic perspective, as well as maybe practically when you’re thinking about implementing way to harvest them?
Roni Israelov 47:01
Yeah, so I think it’s a complicated answer. I mean, at the heart of it, they are different. If you construct a volatility risk premium strategy, you can construct it such that it’s beta neutralized, I mean, typically people will Delta hedge, if you delta hedge, you’re going to be left with some residual beta because of the relationship between changes in implied volatility and underlying equity returns. But you can accommodate that and adjust for it and beta neutralize. That means that it’s possible to construct a VRP strategy that is unconditionally uncorrelated to the underlying market. And that strategy has historically had pretty attractive returns reasonably high returns reasonably high Sharpe ratio. If you calculate a tail risk measure for a reasonably high return relative to the tail risk it brings to the portfolio, which means that it could be an attractive addition to a portfolio. And because it’s beta neutral, you might argue it’s a strong diversifier. The other thing about beta neutral VRP is I mean, okay, by definition, it has no beta, but it has exposure to other things. It has exposure to implied volatility, to realize volatility and other Greeks. So, you know, when you take all of that into account, I think it’s different than allocating to the underlying market. However, here’s my caveat, during market crashes, so during equity market crashes, they both suffer. So they have conditional correlation and correlated downside risk. And I think it can be dangerous to allocate to a volatility risk premium strategy under the underlying assumption that it is a terrific diversifier because it is the example I think of the expression of diversification fails when you need it most. I think sometimes that’s applied to, you know, global equity diversification. And some people might challenge that criticism, but I think that criticism applied to volatility risk premium is a fair criticism, they are by definition conditionally correlated to equities. And I think it’s important to take that into account if one is going to make an allocation to the risk premium. So I think it’s a good diversifier. I mean, it is uncorrelated, it has attractive returns, but it is conditionally correlated. So it’s not a perfect diversifier. And that is important, and I think that is why and part covered call strategies and poor reading strategies are popular, because they essentially do take that into account. Typically, one who allocates to a covered call or a put right is replacing an equity allocation with so they’ve taken a portfolio that has a beta of one and replaced it with a portfolio that has a beta of point five or point six, but in the event of a crash, you know, can have a beta approaching one so you get the conditional correlation, but you’re not actually increasing exposure to equity, downside risk, and I think To those versions of allocations are probably a more responsible way to allocate to the risk premium, practically. And I think this is similar to something that I said, for corporate bonds. But practically, I think there is a natural comfort that people have when allocating to traditional risk premium equity risk premium and bond risk premium and a natural discomfort when allocating to volatility risk premium. So even though it has, you know, diversifying returns, albeit with conditional correlation and attractive, risk adjusted returns, I think most people’s natural inclination is just to avoid it. And that’s just a consideration.
Corey Hoffstein 50:40
One of the practical realities of trying to harvest the VRP is that there’s this reset that happens if I want to capture the equity risk premium, I can just buy equities. And it’s buy and hold. Yeah, I have to take into account the equities that IPO and those that go away, and there’s some rebalancing there. But it’s not the same as say, constantly selling a one month at the money straddle and having to roll that position. I’m curious as to how that sort of rolling reset introduces a potentially compounding effect that needs to be considered with VRP that maybe doesn’t need to be considered with the equity risk premium. Something that comes to mind, for example, is that selling really teeny puts, empirically is had the highest ex ante expected VRP. But ex post, it might not make a great strategy because of that one blow up, destroys your performance forever that Erica diversity issue. Curious how you think about the compounding problem with VRP?
Roni Israelov 51:39
Yeah, so I think there are a number of things to unpack on that. So let’s start with the teeny pots, you know, the deep out of the money puts, or the VRP might be heightened the ex ante VRP might be heightened in those options. And I think there are different ways to measure that one way to think about that is the percentage of option premium retain. So you know, if you sell a deep out of the money put for $1, you might expect to capture 50 cents of that. Whereas if you sell a modestly out of the money put for $1. And maybe you expect to capture 20 cents of that. So that’s one way to describe it in other ways the differential between implied volatility and realized volatility. So when you think about measurements like that, it might be natural to assume that those are more attractive positions to hold in a portfolio. I would challenge that assumption. And I think that’s something that we actually did in a paper called which index options should you sell? This was a paper that I co authored with Harsha Tamala, at AQR. And in that paper, what we did is we looked at the returns of options across the volatility surface long dated options, in short, dated options out of the money, put options, and out of the money call options. And we started by looking at returns, then we looked at return to risk where the risk was a Sharpe ratio only because we think a lot of people do that. And then it was basically warning, volatility is not a good measure of risk when selling options. A better measure of risk is the tail risk, or the crash risk that you’re introducing to a portfolio when selling options. And what we did is applied stress tests, which are pretty similar to the stress test that prime brokers would use. If you have a short option position in your portfolio, where you shock equities down 20%, you shock implied volatility up by a certain amount, and then reevaluate the PnL of the option position under those stress test. And what you find is that if you look at these deep out of the money, put options that have a very attractive Sharpe ratio. And it turns out, they don’t have a very attractive return to amount of stress loss that they introduce into a portfolio. And we think that that is, you know, we thought that that was very important. And I think if you construct a portfolio, where you have a risk budget, measured in terms of how much crash risk you’re willing to include in that portfolio, and you take those facts under consideration, then you are less likely to allocate in general to those Tini posts. So that’s first just a discussion on do we like teeny puts or not? So now let me get to, you know, the question about compounding effects and the like. So what I would say is that I think it’s actually pretty difficult for people to allocate to pure volatility risk premium funds for a couple of reasons. One, if you construct the funds such that it has tolerable tail risk properties, the returns aren’t so high and it is not capital efficient. You have a lot of cash supporting these option positions and people do not like allocating to capital inefficient vehicles. On the other hand, if you want to get the fund to have an attractive return profile as a standalone entity, then you have to size positions to the point that they have potential catastrophic loss risk. And if you’re transparent about that, if you create this fund, and you tell people if equities are down 20%, this fund is going to be down 60% few people will allocate to that vehicle. And I think, you know, when you see some of the blow ups that occurred, let’s say around COVID, I think there were funds that were capital efficient, that maybe had sizable stress exposure. And it’s possible that those who were allocating to them, you know, weren’t fully aware of the amount of tail risk that those funds were providing. That said, so I think I’m making the argument that I think it’s difficult to allocate to a standalone VRP vehicle, that doesn’t mean that it should be difficult to allocate to VRP as a risk premium, it’s just a question of how do you do that. And I think a good candidate for VRP allocations is, as an individual sleeve or multi strat fund, or as an overlay on other exposures. So as an example, imagine you have a bond exposure, a treasury bond exposure, which is not particularly risky, on its own. And you overlay a small modest short volatility allocation. What do you have you have a synthetic corporate bonds with properties that are potentially favorable to actual corporate bonds, that’s a tolerable way to receive that allocation. Or you can imagine a return stacking framework where maybe you’re including some short option exposure on top of 6040, or 9060 portfolio and an integrated portfolio account. So I think there are ways of introducing it. And if you do it in that way, then the amount of tail risk that is being introduced into the portfolio is quite modest by construction, by construction or by design. And because of that, I think the compounding effects that you’re talking about no longer are a concern and the way they would be if the exposures were sized to be very large,
Corey Hoffstein 57:12
I want to jump to a slightly different topic. On this episode. So far, you’ve mentioned something a bit subtly talking about your research, but I know that there’s an approach you like to use. I’ve seen it used multiple times in your papers, where you’d like to work in simulation environments, not just with market data, you’d like to sort of set up these clean lab simulation environments. And you said to me on our pre call you like do this, because it allows you to know where core truth is, and allows you to, then with that core truth, understand the potential marginal benefits of trying to improve models? How good can you actually get versus, say, a crystal ball? It’s wondering if you could explain this idea a little better than I have, and provide an example of maybe ways in which you’ve used it in the past.
Roni Israelov 58:01
Yeah, so I am a fan of simulations. And I mean, they might be half or more of my papers and applying them to help gain insights. So I think they’re useful for a number of reasons. I mean, one, you highlight it, because you know the truth. And it provides a framework to better understand some of the outcomes because you have the advantage of knowing the truth where with real data, you don’t know the truth, you know, part of the exercise is trying to have an estimate of that. A second advantage of simulation is that it allows you to generate many paths, where history gives us one path. And it’s easy to over index on that single path that you realize in a backtest. And I think simulations allow one to kind of correct for that. But you brought up another point, which is it also allows you to test things you really couldn’t test otherwise, which is, you know, suppose you can actually improve a model by getting closer to the truth is that material or not? So if we go back to the idea of the risk modeling exercise, the framework that I was talking about in terms of global asset allocation, was applying the risk model to target volatility. I think I use the example of you want, you know, maybe two or 3% volatility in equity markets. And in order to target that volatility, you have a volatility forecast, it has some estimation error, but you size, the overall position, according to the forecast. And you can keep working on trying to refine that model to you know, have the quality of the forecast improved. But the question is, Are there benefits to doing that? And here’s where I think a simulation approach can help answer that question, because essentially, you can model volatility dynamics in a simulated environment, estimate volatility as you would, as you’ve calibrated it, apply that estimate to your portfolio construction, evaluate the properties of that portfolio, and then you know, the true volatility because it’s part of the simulation. So, you can take a weighted average, essentially Have your estimate of volatility, and the true volatility, essentially cheating, but allowing you to ask the question, what if I reduce estimation error and get closer to the truth? And see, how much benefit do you get from that? If the benefit is modest or immaterial, then there’s little reason to continue down the path of improving the model. I think shot selection as a researcher is very important. We all have limited time, and resources. And we want to apply our time and resources and energy to those projects that are going to have the biggest economic benefit. And I think, oftentimes, part of the research process is actually figuring out an optimal stopping point, that you’ve kind of milked a project for all it’s worth. And it’s time to move on to other projects. And I think oftentimes, simulations can help one ascertain when that stopping point should be, I give one more example, a simulation environment that I’ve applied more recently, because that’s, you know, that’s kind of an older example. But we were recently looking at tax loss harvesting, automated tax loss harvesting. And when you look at different applications of it, there are some that are harvesting losses, you know, on a monthly frequency and others that are harvesting losses on a daily frequency and those that are harvesting daily, you know, advertise the benefits of that. And I think it’s a reasonable question, in terms of like, how much value do you get for moving from a monthly harvesting cycle to a daily harvesting cycle? And I think, you know, intuitively, a lot of people would naturally think that there must be a lot of benefit to that, because the opportunity set is potentially 20 times larger, right, or 20, or 21 times larger. But it’s interesting to actually ask the question. So, you know, in order to ask and answer their question, we built a simulation environment, this was a beast to build, but we were simulating underlying equity returns with dynamics that are realistic in terms of industry risk exposures, and market risk exposures and the like. And we were applying within this simulation environment, a monthly cadence harvesting, and a daily cadence harvesting and looking at the benefits and cost of different thresholds for harvesting. And there’s a trade off at play. And this is just natural and so many different processes that you have to trade off a benefit for a cost. With harvesting, the benefit is that you’re able to harvest a loss, the cost is that you’re taking active risk relative to the desired position, because there are wash sell restrictions in play. And once you harvest a loss, you can own the position for 30 days. So if you’re aggressive in harvesting losses, you get more tax loss, harvesting yield, but you’re increasing the risk exposure, you know, the tracking error of the portfolio. So it’s a traditional, like mean variance frontier, to some extent. And essentially, you can build that friends here, under daily harvesting regime, build the same friends here on a monthly harvesting regime and look at the differences. And what you find is pretty interesting. If you establish the same threshold, let’s say a 10% threshold, harvesting daily yields more losses than harvesting monthly, there’s no surprise, but harvesting daily leads to more trackier than harvesting monthly. So it’s not an apples to apples comparison to just say that you get a greater harvesting yield. When you move to daily. The question is, can you move the frontier by moving from monthly to daily? Can I get the same harvesting yield with lower tracking error? Or can I get the same tracking error with a higher harvesting yield? And what that analysis showed is those frontiers are nearly on top of each other, that there’s actually very little benefit to moving from a monthly harvesting cycle to a daily harvesting cycle. And using simulations. And having some control over the underlying environment allows us to answer questions like that, which I find to be pretty helpful. Well, it’s
Corey Hoffstein 1:03:57
a great teed me up nicely to segue here, because that’s a whole different style of research. And I think it speaks to where your role is today at Endeavor, which is a financial advisory firm for high net worth individuals. And I’d love to know from your perspective and making that transition over the last couple of years how has the move from biocide asset management to financial planning changed? Sort of the core research problem you’ve been focused on?
Roni Israelov 1:04:24
Yeah, so I think the first thing I would say on that is I think it’s expanded the research problem because for our organization, we provide advice and planning services, but we also construct custom portfolios, goal oriented portfolios for our clients. So we continue to also engage in the type of research you would see an asset management. So I think our research tends to be comprehensive and we’re just kind of active across the board. In terms of the production of research, I think it’s pretty similar. I mean, we have a strong research team made up of PhDs and Masters of Science in computational finance. Our decisions and algorithms rely on and require rigorous research. So everything is well motivated. We’re researching factors and portfolio construction, rebalancing, and so on. So there are a lot of areas that are pretty similar. I think one point of difference is not on the production of research, like what we choose to work on, or how, but the consumption of research, you know, I think when you’re in an institutional asset manager, a lot of the consumers of research are finance professionals. So, you know, you’re kind of speaking to an audience of peers. And that research where I think, as an RIA, some of the consumers of research are kind of pure as finance professionals, but other consumers are not necessarily finance professionals. So it’s a different audience. So that would be one comment. But to the other comment, I think it kind of expands the aperture for the type of research that we do. And on that, I think I would give two examples. So the first example is that part of the service that we provide is financial planning. And that requires rigorous analysis, you know, custom analysis for each client and their plan. And in order to assist our clients in answering questions. And given what we just talked about, this may not be a surprise, but we focused a lot of attention on Monte Carlo simulations, building sophisticated, accurate Monte Carlo simulations. And this was a huge project. And I mean, I could probably spend an entire hour and a half, you know, or more talking about this. But let me let me just speak to one example. And this is something that Stephanie Lowe, one of our researchers spend a lot of time on. So one of the strategies that we make available that I think is pretty unique in our space is liability driven investing. Our clients have cashflow goals. And one way to de risk those cashflow goals is to immunize them against changes in yields by building duration match bonds. And if we want to help them understand whether it makes sense to do that or not, or to what extent it makes sense to do that, you know, what are the trade offs that they’re making in terms of returns versus their portfolio risk, we need an accurate way to simulate the behavior of their portfolio when implementing a strategy like that. And in order to do that, we had to simulate yield curves. So we built a yield curve simulation model where we’re simulating level slope and curvature. And we wanted it to be an accurate representation of bond behavior. So we use the simulated yield curves to price bonds, existing bonds and hypothetical bonds in the future. And we were looking to calibrate it such that it had properties that are consistent with what we see in bond markets. And by that I mean, if you look at bonds returns tend to increase in duration. Volatility tends to increase in duration. But Sharpe ratio tends to decrease in duration. And you know, this was empirically demonstrated by I think, Andrea for Xenia and Laci Peterson and a paper at AQR, but those are, you know, some properties and perhaps due to leverage aversion that you see in bond markets? Well, if we want to accurately represent the behavior of these portfolios, we want to make sure that our simulations have those properties. And they do and I think, you know, the returns and volatilities and Sharpe ratios are reasonable. Another reason why it was important in this example, to be able to simulate yield curves and the behavior is because you might have bonds for different purposes, you might have bonds for the purpose of immunizing future cash flows as a risk reduction technique, you might also have bonds for the purpose of diversifying equities and a growth portfolio, you know, in this synthetic corporate bond type of portfolio, and you want to be able to understand the total risk and return behavior of the portfolio. So for that, again, it’s important to have accurate simulations. So we’ve devoted a lot of research resources to building a Monte Carlo simulation environment and analysis engine. And I think that is something you wouldn’t typically find that an asset manager, I would argue you’re not going to find this at another financial planning firm either, but it’s something you know, that is unique to endeavour. A second point that I would highlight is on the portfolio construction, asset managers, you know, talented asset managers are going to focus a lot of attention on portfolio construction, we’re doing the same, but the nature of portfolio construction is pretty different. If you’re a large institutional asset manager, you’re probably managing a small collection of very large commingled vehicles. And a lot of attention is paid to try to get as much juice out of those portfolios as possible, increase the return have a better risk model. Think about modeling trading costs, how you execute it, and you know, you’re kind of doing everything you should do as a responsible asset manager, but you’re managing, you know, a 1 billion 5 billion $10 billion Are commingled vehicle. For us. Our goal is to manage 1000s, maybe 10s of 1000s of custom small portfolios, million dollar portfolios $2 million portfolios. And that requires an entirely different portfolio construction framework, I do not think that the traditional institutional portfolio construction framework would scale in a way that could be applied to this market. So we have had to do considerable research and essentially devising new, robust portfolio construction methods that would support the building of custom plan oriented portfolios. For a large set of clients.
Corey Hoffstein 1:10:44
One of the things I want to return to is this idea of risk, we spent a lot of time talking about risk from the buy side asset allocation perspective, I want to talk about risk from the financial planning perspective, you mentioned to me in our pre call, that you end up looking at risk across two different vectors for your clients, I think what you call the risk to capital, and the risk of people being able to meet their objectives. And one of the things you said that I thought was really interesting that I’d love for you to spend some time expounding upon is that sometimes a portfolio solution can work really well, along one of these risk factors, but not the other. And that actually simultaneously solving for both can be really difficult. And I thought that was, that was really interesting. I was hoping you could expand upon that for me.
Roni Israelov 1:11:32
Yeah, so I think wealth optimization is complicated, you know, which is what we’re trying to do. If you go back to traditional mean, variance optimization, what is the framework, you like returns, you don’t like risk, your model risk is variance, that’s mean, variance optimization. And then if you can figure out some risk aversion for a mean variance investor, you can find an optimal portfolio. And there are questions about, you know, how do you identify this risk aversion and that in and of itself is complicated. But I think when you go to individual investors, where they’re complex goals, it becomes even more complicated because I understand why a version of variance, you know, which is capturing risks to capital matters, someone invest a million dollars, they don’t want to lose a considerable portion of that. But for many investors, their wealth is in service to longer term cashflow goals, retirement education spend for their children, you know, whatever it might be. And it may very well be the case, then in order to meet those long term cashflow goals, risk is required, right, if you don’t take risks to capital, you are taking risk in terms of your ability to meet those long term capital goals. And I think the simplest way to demonstrate this is under a liability driven investing framework. So let me just walk through that for a moment. Imagine you have a goal, whereby you need $100,000, in 30 years, if we want to minimize the risk of your ability to meet that goal. And this is something that’s you know, taught in most finance classes or investing classes, but you can buy a zero coupon bonds, right, you buy a 30 year zero coupon bond with a face value of $100,000. Let’s take taxes out of the equation for a moment. And maybe that costs you I don’t know, 20 or $30,000 today, and you invest that amount of capital today and 30 years, you get your 100 grand, and there is no uncertainty about meeting that cash flow goal. However, a zero coupon bond has a ton of volatility, a 30 year zero coupon bond is in all likelihood more risky in terms of volatility than equities. And anyone who wants to think about that for a moment can just look at bond returns last year, and consider the risk that bonds brought into individual portfolios. So I think that highlights the trade off, I mean, that simple example, highlights the trade off at play. And I can just expand it that idea for a second. But if someone is entirely concerned about risk to capital, then they can keep their money in a cash account, earn whatever cash they earn over the 30 year period. But there’s tremendous risk because of you know reinvestment risk in terms of whether they’re going to be able to meet their cash flow goal. On the other hand, if they’re concerned about their cash flow goal, they can invest in a zero coupon bond and take all of the capital risk that is associated with that. I don’t think there’s any right or wrong answer. And that’s what makes this challenging people have different risk preferences. So one person may care more about risks to capital than another person who cares more about plan risk. And I think what’s important is being able to provide the information that someone needs to make an informed decision and allow them to provide us with the information we need to help them To make an informed decision, and I think this idea that complexity of it is one problem with traditional risk tolerance questionnaires is they seem very focused on the risk to capital and less focused on this other important risk, which is the risk to being able to meet someone’s plan
Corey Hoffstein 1:15:21
goals. You recently published a piece titled, how many stocks should you own, which was a push back on the common idea that you see sort of thrown around in, in even like finance 101 classes that the marginal benefit of diversification rapidly to clays after 20 or 30 stocks? Once you get beyond that portfolio volatility tends to die down? Your pushback in this piece was that those numbers are too low, potentially off by an entire order of magnitude? Can you walk me through that?
Roni Israelov 1:15:55
Yes. And what’s interesting about this is if you Google the question, you know, how many stocks should you own? And now you can ask Chet GPT? The same question, you know, I guess we have expanded sources of information. The response of 20 to 30, is very common, and I think they generally mostly tie back to the same original study by I think, are turned evidence, but the answer is based off of an estimate of the volatility of a portfolio. So the typical approach one would take to derive that answer is you have a very simplistic model of equity, volatility is whatever it might be 15%. And stocks have a certain amount of idiosyncratic or name specific risk 30 or 45%, whatever your number is, and as you allocate to more stocks, you get tremendous diversification on the idiosyncratic risk, and the total volatility of the portfolio quickly converges to nearly it’s lower bound. And when you look at the curve, it’s hard to challenge. I mean, if that’s your framework, it’s hard to challenge the conclusion of 20 to 30 stocks, because the volatility is not materially higher than the volatility of a 200 stock portfolio. I think the issue that we had when we pursued this, and this is work with Yan Chen, one of the researchers at endeavor, but I think the issue we had is there’s an underlying assumption in that model, essentially, that all these stocks have the same expected returns. So by picking a random set of 20 stocks, you’re not changing the expected return of the portfolio, the only thing impacted is the volatility. And I don’t think that underlying assumption holds, I think different stocks have different expected returns, we may not know what they are, I mean, you can have some models for that based off of factor exposures and otherwise, but you know, their actual expected returns might be unknown. And if you look at the distribution of returns, you actually see quite a bit of variability. So what we did is, instead of using volatility as a proxy, for bad luck, what we did is looked at the terminal wealth of a portfolio as it relates to the number of names held. And we picked a number of random portfolios, I think, 1000 Random portfolios, for each number of names, 2550 75 100, and so on and so forth, looked at the terminal wealth, looked at the 10th percentile and 25th percentile of terminal wealth, to try to see how bad is it if you get unlucky? And what if you get like a one in 10, unlucky outcome, how much of the expected terminal wealth are you foregoing? And what we found is if you have a 25 stock portfolio, the unlucky investor, this is a 10th percentile unlucky investor ended up forfeiting 36% of their expected and wealth, but a 250 stock investor only forfeited 10% of their expected on wealth. And if we compare these two investors to each other, that’s you on your 50 stock investor had a 40% higher and wealth than the 25 stock investor, you know, when we’re comparing the unlucky investor to the unlucky investor. So to us that signals that there are still material benefits to diversification beyond owning 25 names.
Corey Hoffstein 1:19:09
There you received, what I thought was some surprisingly strong pushback on this article. I love the article. It’s something that I have studied for years, this idea of terminal wealth, dispersion versus volatility. But there was you put it out on Twitter. And there was a lot of people who just completely disagreed with the precedent, and then actually sent you back to the lab to do a little bit more research. I think some of the pushback maybe was reasonable, maybe introduced some other questions. I do know you’re planning on publishing a follow up, which may or may not be out by the time this episode airs. Maybe you can talk a little bit about the critiques and what your follow up research has found.
Roni Israelov 1:19:49
Yes, so the primary critique is related to the fact that when we randomly selected stocks, we pretty much did it in the most not each way possible, we literally randomly selected stocks. And what many people said is that was too naive that it would make sense. And any reasonable person would do this, they would set your stratify when holding a, let’s call it low named portfolio. And in fact, if you go back to some of these Google searches and look at some of the recommendations, you know, I don’t know maybe half of them, when they suggest 20 stocks will say something like, and be careful to have, you know, maybe two stocks from every sector, so that you are appropriately diversified. And I think, you know, those of us that are familiar with equities, well understand that sector risk is a real thing. I don’t challenge that sector risk is real. But I think what’s misunderstood is the amount of idiosyncratic risk relative to Sector risk, that sector risk is, is real, it’s important. But if you own 20 stocks, the portfolio is still being dominated by idiosyncratic risk. And in order to demonstrate that, essentially, we just repeated the analysis with a sector stratification people were hoping for, and their criticism. So we pluck stocks randomly, you know, 2550 100, names, whatever the case may be, stratify those stocks across sectors, and then look at the outcomes. And what you find is a slight reduction in terminal wealth dispersion for the sector stratified versus the non sector stratified, that’s to be expected, because you’ve gotten rid of some of this risk, but the core finding is unchanged. If you look at the plot of terminal wealth dispersion or 10th percentile outcome as it relates to the number of names held in the portfolio, when you sector stratify, you find the same thing that the unlucky 25 stock investor whose sector stratified their portfolio has materially less terminal wealth than the unlucky 250 stock investor, who also sector stratified their portfolio. So the finding is basically, sector stratification reduces risk a little bit sector stratification is no substitute for diversification.
Corey Hoffstein 1:22:13
So as a man of research, I don’t think people will be surprised to find out that you like to research and experiment in your life outside of finance, as well. And I have it on good authority that you like to run a lot of experiments on dry aging beef. Yes. So as we come to the end of the podcast here, any tips or tricks for myself and the listeners?
Roni Israelov 1:22:36
I think it’s mostly experimentation. So I got into this after I had visited Budapest, we have an office in Budapest, I got a bit of pest and there’s a surprising number of restaurants that offer dry aged beef, and I liked it a lot. I mean, I’d had it in the US before, and I decided I wanted to do this on my own. But I didn’t know I mean, there are a lot of combinations, you know, a lot of parameters, if you will, in terms of how you can dry aged beef, you know, do you wager for 30 days, 60 days, 90 days? Are you using prime beef, select beef choice beef, which cut you know, New York Strip ribeye, I mean, there are a lot of lot of degrees of freedom here. And the only way to figure out preferences through experimentation. So I’ve been running experiments, I think preferences here are personal just like they are in other areas of our life. I have mostly converged on 30 day, dry age ribeye or New York Strip, I think I have a slight preference to ribeye, but they’re both quite tasty, to me 60 days beyond the flavor profile that I’d hoped for 90 days, didn’t like it at all. Also tried for dry aging grass fed beef, I wouldn’t call that necessarily a success, but you don’t know until you try. So I think it’s just an area of experimentation. Unfortunately, for me, neither my kids nor my wife likes dry aged beef. So I’m kind of on my own. With this. That’s one problem. And then the second problem I don’t know if people are familiar with this, but when you dry aged beef, you have to cut around the out, you cut the outside off because it becomes crusty and, and dry and it doesn’t impart a nice flavor profile to the steak. And that’s called pellicle. And apparently, you can use that pellicle for something I have not figured out how to use it effectively. So I have now collected, I don’t know one or 200 pounds of pellicle on my freezer. So if anybody who’s listening to this podcast has any ideas for me on what to do with this pellicle I would be most appreciative sure
Corey Hoffstein 1:24:30
your wife loves that last question for you. So as listeners know as they look at these episodes, the artwork for this season is inspired by tarot cards that I’m asking each guests to pick the tarot card they want as the design of their cover. You chose the ace of swords, and I know you weren’t familiar with Tarot cards, nor did I expect any guests to be nor am I particularly familiar with them, but I thought it was a cool design and an interesting opportunity for people to pick something that might resonate with them. So My final question to you is, why did you pick the ace of sorts?
Roni Israelov 1:25:04
Yeah. So as I was looking at this, I noticed and I had no familiarity with Tarot cards before this question, but apparently, they are supposed to tell you something about your personal or family life or love life, your financial life and your career life. And I had a hard time finding one that accurately represented all those dimensions that was very challenging. So what I decided to do was keep my personal life personal and focus on the tarot card that I thought maybe better described my professional experience. And I felt like this connected well, with my experience of having joint endeavors. So for those who are unfamiliar with the Ace of Swords, which I’m going to assume as most people, my reading of it is the following The Ace of Swords indicates that someone is about to experience a moment of breakthrough, and that would be great. It’s associated with keywords such as clarity, new ideas, concentration, focus, and truth. And those keywords resonated with me, with respect to career, it represents a new career that can bring a lot of intellectual stimulation is delightfully challenging, offers encouragement to grow mind and skills and surrounded by colleagues that enjoy discussing new ideas. I think that’s a nice reflection of my time at Endeavor. It’s maybe not new anymore. I’ve been here for three years. And then lastly, and maybe this is good advice for people. But with respect to finance, it says trust your brain and avoid making emotional decisions and it offers a warning rejecting the rational decision might feel good, but it’s costly in the long run. Ace of Swords. Roni, this
Corey Hoffstein 1:26:38
has been fantastic. Thank you so much for joining me.
Roni Israelov 1:26:41
Yes, thanks for having me.