Trading Tomorrow - Navigating Trends in Capital Markets

The Potential Future of Portfolio Management with Agentic AI

• Numerix • Season 4 • Episode 39

Share episode

With volatility at levels high and risks evolving faster than some team's action plans, investment professionals may need a new approach. In this episode of Trading Tomorrow - Navigating Trends in Capital Markets, host Jim Jockle is joined by Rajiv Bhat, CEO and Co-Founder of Martini.ai, to explore how AI-powered scenario modeling is transforming credit risk management. From real-time portfolio stress testing to the rise of agentic AI as your investment co-pilot, discover how to turn uncertainty into strategy and resilience. 

Jim:

Welcome to Trading Tomorrow Navigating Trends in Capital Markets the podcast where we deep dive into technologies reshaping the world of capital markets. I'm your host, jim Jockle, a veteran of the finance industry with a passion for the complexities of financial technologies and market trends. In each episode, we'll explore the cutting-edge trends, tools and strategies driving today's financial landscapes and paving the way for the future. With the finance industry at a pivotal point, influenced by groundbreaking innovations, it's more crucial than ever to understand how these technological advancements interact with market dynamics. Interact with market dynamics.

Jim:

Today's episode dives into urgent questions facing investment professionals today. How do you build a resilient portfolio where there's so much instability in the world? From interest rate shocks and geopolitical tensions to the ripple effects of economic policy shifts? Today's market volatility demands the latest and greatest tools. Now the question is how can AI help? Joining us to explore the question further is Rajiv Bhatt, ceo and co-founder of Martini AI, a leader in AI-driven credit analytics. Rajiv brings a unique blend of technical depth and entrepreneurial experience to the table. He holds a PhD in theoretical physics, previously led data science at AdTech, unicorn in Mobi, and founded Y Combinator, a startup acquired by Groupon. At Martini AI, rajiv is helping financial professionals rethink portfolio credit risk management through intelligent automation, real-time credit signals and scenario modeling that captures the interconnectedness of today's macro forces. Rajiv, thanks for being here.

Rajiv:

Thanks for having me here, jim, very excited, big fan of your work and always excited to be up close.

Jim:

Well, you know what? Why don't we jump in? We'll start with the basics. So what exactly is AI-powered scenario building and why is it so important in today's investment climate?

Rajiv:

The world's been changing much faster than it was before and it's also much more connected than it ever was before, and what that means for an investor is that cumulative risks are now just getting more and more pronounced.

Rajiv:

So you need to be like situations that you could never have imagined can now suddenly arise and can kind of run away from you, and which is why we are very excited about how you can grapple with that beast by using AI. So that's what the AI-powered scenario builder is. It's a way in which you can understand the fast-changing world around you as it impacts your portfolio. So that's what we built. So it's a very simple tool where you just type in what do you think is going to happen, or maybe the Panama Canal getting a shutdown and you can see what happens to your credit portfolio. It could be a portfolio of thousands of companies and it'll tell you what are the expected impacts, how does the expected loss change, what is the value at risk, which companies get impacted and this is all quantitatively estimated and it gives you that in seconds.

Jim:

And so how would you categorize the volatility we're experiencing today versus past cycles and, from your perspective, what does that really mean for risk?

Rajiv:

management.

Rajiv:

The volatility that we are seeing now is matching what we were seeing during the GFC before.

Rajiv:

It's very heightened because of several things like geopolitics, the tariff situation that's going back and forth.

Rajiv:

Folks are unable to figure out how some of these fast-changing events are going to impact their historically slower-moving portfolios, and I think a couple of things that it's impacting managers is. One is in terms of understanding what's the risk in their portfolio. Like, though the assets are illiquid, what happens to the risk? Because the risk is still liquid, like you know, your companies are still going to get in trouble, though you might not be able to sell the debt. So that's one way in which it's impacting portfolio managers, because it's gotten them thinking about how do I handle the situation now. Like you know, historically the kind of actions that they could take were commensurate with the timescale of, like, the timescale of the risks that were unfolding, but now what's happened is the risks have gotten ahead of the timescales of action, so folks have fewer options to respond to the risks that they have, and people are beginning to think about how do I operate in this new world, and they're thinking of new tools, new kinds of strategies, and that's gotten everyone thinking about it.

Jim:

So how can AI help managers turn that uncertainty into a more strategic advantage?

Rajiv:

That's such a great question and I think that's a question that's not asked enough really which is saying AI. There's a fair bit of commentary on how AI can help you save time, can make your things, can make work more efficient, but the real question people should be asking is how can AI help me operate now? And, like you rightly said, how can AI take me to the next level, Like how can give me tools that are going to be superior in the new world? Two ways. One is that AI is phenomenal for operating with large volumes of data, fast-moving data, making sense of what's happening right now, for example, the tools that I was just mentioning, the AI-powered scenario builder. It does exactly that. What it does is it looks at how every company is situated in the universe. Every single company is calibrated to different risk factors. It throws all of that together and it figures out OK, this is what's going to happen to this portfolio if this kind of event manifests itself, and so that's something that would take analysts four weeks of time to do.

Rajiv:

Now it takes a second, and so that's the first thing, which is AI helps you just get on top of the existing situation. So now, if you're a portfolio manager, the next thing you do is well, that's great. I can do this in a few seconds. How about I do this for 100 different scenarios and let's see what happens to my portfolio, which companies are the most vulnerable, which companies are going to thrive in these new situations? Let me get into that. It makes for more well-rounded decisioning there. The second thing is to say can AI come up with better strategies? And that's the new, powerful thing that's beginning to make its way up, which is helping using AI to design policies, to design strategies. To say, in a world where how should I design my portfolio, how should I structure my portfolio? What assets will make sense in these scenarios and what should me buy off ramps in case some of these situations don't pan out? Just being able to have a more thorough plan which is in far more fine structure than ever before? I think that's what AI lets you do.

Jim:

I've heard excuse me, I've heard stories where sometimes you need to blindly trust the AI. Right when you're asking for scenarios, especially around portfolio management, perhaps the AI is finding different patterns and things of that nature and that, to the extent you're asking these tools to make things or the logic human, understandable, sometimes that deprecates performance. Have you found that to be true? Oh, it's such a great question.

Rajiv:

If you the temptation is there, just blindly trust AI and let it just follow the overlord and do what it's asking you to do, and to your point. Yes, if you do that, you will run into some of the blind spots that AI will have right, like there are, but the reason is basically this right, which is that we live in a very high dimensional world, like everything that we do has so many aspects to it and it's very fine textured and there's only so much information you can communicate in a prompt. And so an AI is going to make some assumptions, and if you blindly trust it, you're assuming that its assumptions are the assumptions that you have, which is not often the case, and so, unless you are doing a process where you're double-checking you're validating what its output you will quickly run into a situation where what the AI is thinking in its head is different from what you have in mind. So you're validating what its output. You will quickly run into a situation where what the AI is thinking in its head is different from what you have in mind and it's not going to line up well, and to that extent, the way we think about it is we think of the tools that we are building, particularly the scenario build, as well as the Ironman suit.

Rajiv:

It's not Ironman, so it's basically letting you do so many things, even in our scenario builder. For example, if you have a scenario where invades Taiwan, if our AI comes back and says, you know, I think the S&P is kind of going to crash about 10% because of the pressure of the semiconductor industry, it lets the user exercise editorial oversight, come and move the slider down to say 80%, if that's what they think is going to happen, and then look at the scenario. So our design principle there is have the output of AI. It's always going to be verifiable and changeable so that you can do your work fast. But it doesn't mean that you let go of the wheel and let AI do the driving.

Jim:

So what does scenario building look like in practice? Can you walk us through an example, sure?

Rajiv:

In our case, the way we think about credit portfolios is in three ways. There's companies, portfolios and sectors, and our scenario builder operates when you have a portfolio loaded, so a portfolio with essentially a list of companies and with some percentage allocations to each of them, and this is where the magic starts. So once you have the portfolio loaded up, building the scenario for us is just typing it in, so it's like you just type in the scenario that you think you want to test your portfolio against. So it could be something like you know what happens if tariffs on imported metals goes up 100%. Or you can say there's some geopolitical conflict in Europe, or you could talk about a local election and see what happens. So then what? The uh, the ai, takes over from there, where what it does is. It takes that scenario and breaks it down and it says, okay, let me see in the past what's happened with this. Like if we're going back to our example of . It says, okay, let me look at some of the bigger factors this the s&p. There's a price of oil this dollars, this dollars. There's inflation what happens to each of these typical insert scenarios? And then it figures out, comes back with reasonably good estimates on how each of those would move. And that's going back to the previous question. That's where the first bit of editorial oversight comes in. The user can then kind of eyeball those and say, okay, this is in the ballpark or could get better or worse, and make adjustments if required.

Rajiv:

And then what it does is it takes that and then it applies it to all the companies in the portfolio through our knowledge graph. So the way our system operates is that we have we cover three and a half million companies. Each of those companies are embedded in a knowledge graph. The knowledge graph is terabytes of data. Every company is connected to over 100 other companies along multiple dimensions things like you know, transactions, supply chain news, investors, things like transactions, supply chain news, investors. And then we also pull in about 10,000 to 20,000 risk reference points every day.

Rajiv:

So these could be things like prices of bonds, prices of loans, bankruptcies, stock prices and the movement of those, and portfolio valuations and things like that.

Rajiv:

And so because of that, we get a real-time view on risk in the market and how it pertains to companies in the knowledge graph and over the past few years, like we've seen how the spreads and probabilities of default of companies in the graph has responded to all those different events and all those different market events. So because of that now for every single company in that universe 3.5 million companies we have a quantitative understanding of how it moves to different fluctuations in the environment. And so now when we have this top-level situation and it's broken down into all of these different factors, we can trickle that down to each company and kind of figure out how each company will respond to that. And then we roll that up at the portfolio level and figure out what happens to each portfolio response to this. All of this happens in like seconds but makes for a very exciting kind of use case where you can actually play out different scenarios, kind of design, and shape your portfolio accordingly.

Jim:

Yeah, you often talk about building resilience into investment strategies.

Rajiv:

What does that look like, and how does AI enable that? In my mind, resilience in an investment strategy is its ability to withstand shocks, and what level of shocks can it withstand? And, of course, all portfolios are subject to some level of weakness in the face of some shocks, but, in general, your portfolio is better if the kind of shocks that can withstand are higher, and so then that's what I mean by resilience. So now, if you can design your portfolio such that you take run stories from last three months through your portfolio and you see what happens to which are the weakest companies in that portfolio, like which companies are consistently showing up as being impacted by these stories, then, as a portfolio manager, like a simple step for you would be to just drop those companies and then automatically make your portfolio a little better. The other way that you could make your portfolio strong, of course, is through classic hedging mechanisms, so you could design your portfolio such that and that's something that we provide at Martinet as well, for like you can see how exposed your portfolio is along multiple dimensions. The utterly truth is like all of us end up becoming factor investors, whether we like it or not, and so you can kind of identify which kind of factors you're beginning to lean towards and compensate for that. That makes for greater reliance too. And the third way in which you can kind of identify which kind of factors you're beginning to lean towards and compensate for that, that makes for greater reliance too.

Rajiv:

And the third way in which you can beat reliance is this thing that I keep talking about, which is that just because your assets are illiquid does not mean your risks are illiquid.

Rajiv:

Your risks are just as liquid as the stock market.

Rajiv:

And so just being able to and if you try and like I've heard often heard that why would I care about the risk in my private credit portfolio? Because I can't sell it anyway, because you know have to mark the market and you can hold onto it for longer and my response to that always is look, in a private credit scenario, if you're going to be compensating for risk or compensating for some change in your portfolio based on actuals, it's way too late. Like if a company is already bankrupt and you're trying to offload their debt, you're not going to have much luck. On the other hand, if you are able to be on top of it on a daily basis and make small adjustments here and there change the size of portfolio or change the drawdown limits, add in a few more covenants, you could just pick the phone up, call the borrower and ask him what's up. Doing small things like that can suddenly change the shape and nature of your portfolio and so, just understanding that because the risks are liquid, you can adopt more frequent interventions to compensate for that.

Jim:

So what's staying on private credit? Obviously it is booming right now. You know it's the next horizon, I think, although we've seen some shocks amid the trade wars and tariffs at this point in time. But I guess the question is and one of the concerns I've always had is data availability. You know, finding data on private companies can be very, very challenging. So you know, how do you manage private credit within your system, and is there a role for more agentic AI to help improve transparency in that space?

Rajiv:

Absolutely such a great question. So I think that's been one of the most exciting propositions, value propositions for Martini, which is our ability to provide coverage for private companies. So, of the three and a half billion companies we cover, almost all of them are private, and the way we've gotten around that is by stepping back a little bit and saying, hey look, in this day and age, every meaningful company has to have a significant presence, digital presence, and it's not just a website, but it's just data sets which are out there. There's structured data sets about, like supply chain, employees leaving joining. There's structured data sets around, like you know traffic, investments, transactions, news and it's an elf, it's an enormous treasure trove out there. And so then, the question that we wanted to answer was saying, given that there is so much structured and unstructured data out there it could be filings, it could be data sets, it could be news what's a good way to set it up so that you can consume everything that's out there? It could be partial, noisy, incomplete, it could be difficult to parse, and how do you take all of that unstructured data or structured data, incomplete data, and combine it with the fact that there is so much risk signal available in the market by way of all the bond prices Like if a bond trades at 110 instead of 100, it's telling you something about that company. It's also telling you something like companies like it. It's also telling you something about industry in general. How do you combine all of that information and make sense of it all in a real-time way? And that's where I think some of our backgrounds helped a lot.

Rajiv:

So me and my co-founder both have a huge background, like very long experience with machine learning and large data sets. Co-founder Rohit, master's from Stanford, phd from MIT, worked as a PhD at the AI lab at MIT, won his work on knowledge graphs, won the Test of Time Award. Then he ran a hedge fund with the co-founders of Akamai, doing trading up to $3 million a day. So significant experience putting money where his elbows are. My background is I'm a physicist. I did my PhD in quantum mechanics. I'm privileged to work with a couple of Nobel laureates. I've been working on the startup side with big data companies. I'm privileged to work with a couple of Nobel laureates. I've been working on the startup side with big data companies. In my last gig I was working at Intercon.

Rajiv:

We were processing petabytes of data, doing 600 billion predictions a day, 12 billion auctions a day. So we brought a lot of that in, which is basically big data, knowledge graphs, ai, and so then what we did is first we built a huge knowledge graph of companies connected to every other company. The second, we pulled in every single risk market or market risk metric out there. Then we did a little bit of work, which is pretty tough and heavy, which is taking each of the signals and calibrating it to probabilities of default and risk of each of these companies. And then we have these graph neural networks running on these, on the graph, propagating those risk information. So it's almost like saying you have the surface of companies. If there's a disturbance somewhere, how does it propagate? Which companies get impacted?

Rajiv:

And what we found is this we've been very excited about the results. Like you know, all of this stuff can be back-tested and we've been loving the performance of it so far. We find over 80% of defaults happened, the worst 20% of our predictions and we find that we're able to catch signals way ahead. Like in a back-test with a bank, we found we were, on average, seven months ahead of their first sign of non-accrual. So we've been excited about how well it works. And I think it works well for a few simple reasons. One is the quantity that we're predicting is probability of default, which is just like is the company getting better or worse? And that is a nice, well-defined physical quantity to predict. We're not trying to predict alpha, we're not trying to predict stock price. It's just saying is this company getting better or worse? The second thing, the reason it works is no, it's very tough for companies to escape the pressures impacting their sector. Like if Carvana is struggling to sell cars online, it's very unlikely that CarMax and CarGurus are also not feeling it, and I think that's part of the reason. That's something that you'll never catch in financials. And the other reason, a couple of reasons we've been excited about. Zendaya approaches One to your point about data being difficult for private companies.

Rajiv:

I think most people are talking about financials and over time, what we've realized is financials themselves, while they're important, don't are difficult, because one they can be pretty late. It can be up to six months before you get financials for a company quarterly financials. Second, every term in a financial report is defined differently. Every term might be something for one company, it might be something else for another company. They're not standardized, and so for you to understand what each one means is a fair bit of work, and if you have a portfolio of thousands of companies, it's going to be pretty difficult.

Rajiv:

And the third reason is every financial can come in at a different point in time. So if you're trying to sit down on a Monday morning and saying, hey, how's my portfolio risk looking at, you'll probably look at it and be like, hey, you know what? I have data for only like 25% of my companies and I don't have data for the rest. I need to call them up and pull. That. It's pretty heavy work and with our systems, you can actually just, on a daily basis, monitor your entire portfolio. Be on top of that and the go ahead I'll just make one last one.

Rajiv:

Sorry, go ahead, jim, please yeah.

Jim:

No, no yeah.

Rajiv:

I was just going to say that the one thing that we're struck is that every six to 12 months, what we're seeing is most companies are impacted by some unexpected shock to their margin structure or revenue structure because something like you know, raw metal prices going up or tariffs or some competitor coming and eating their lunch and if, like we find, the traditional methods are too slow to catch those things.

Jim:

I guess you know one of the things that just kind of popped into my mind and you know I look at my iPhone, you know, and you know it's going to tell me it's going to rain here in. You know the. You know it's going to tell me it's going to rain here and you know the next 25 minutes. But obviously that's a statistical kind of analysis versus a longer radar perspective. You know what is the time horizon and or is there accuracy degradation over time horizons within within the AI, or is it? You know, like how? How is the AI? How is the AI looking at the universe? Is it a smaller window? Is it longer? Where is the accuracy?

Rajiv:

That's such a great question and the way we do it is at the risk of maybe giving away too much, but the way we do it is we actually look at all-time horizons. So in all debt instruments you have an yield curve and you have like risk at pricing at like a three-month horizon or a one-month horizon. You're like, if you look at the treasury yields, you have yields which are like for a 10-year treasury bond or a one-year treasury bond. It's telling you different things about what the market thinks is the risk at those horizons. So we actually pull that in for as many debt instruments that we can get and we fit everything to that. So for any company that we cover, we actually provide our sense for probability of default at like a three-month horizon or a one-year horizon or a five-year horizon, and we fit everything to that.

Rajiv:

And so what that lets you do is answer this question in different ways. So if you say, hey, give it to me at a one-year horizon, which is what typically everybody looks at, we can do that. And so what that lets you do is answer this question in different ways. So if you say, hey, give it to me at a one-year horizon, which is what typically everybody looks at, we can do that. But if you want to look at a longer horizon, it can do that. So, yeah, so, and it actually helps, because once you are willing to look at all-time horizons, you get a lot more signals. If you try to constrain yourself just to data points at a three-month or a one-year horizon, it's not enough.

Jim:

You know. One of the things that you're making me think about is you know you have so much market observable opinion coming out of the market, whether it's bond spreads or credit default swaps, et cetera, and you have so much data and information being produced by organizations. What is the future role of the rating agencies given this new AI-related world?

Rajiv:

Absolutely, jim. I think what's taken everyone by surprise is the rate and speed with which AI can now process unstructured data and illiquid data and sparse data, which is exactly what this entire rating industry works with. And from the rating agency's perspective, I think it's inevitable. I feel because in our signals we see that we are frequently six to eight months ahead of any rating changes, because just the markets know, and just being able to pick that up and propagate it itself already makes it very useful. Uh, from a to your question about what's the role of rating agencies going forward, I think it's rating agencies owe it to make the latest and most important ratings available to the users or anyone who needs it. In fact, we are beginning to think a little more broadly and we feel maybe we should be available to everybody, anyone who is operating in the space. Like you know, earlier data was free. Now, maybe, going forward, maybe insight will be free or opinions will be free, and so in that world world, how does things change? Like uh, and I think, uh, if I were. Uh. So, in fact, we don't think of ourselves. We don't think of ourselves as a rating agency. We think of ourselves as a credit interpolation company, and what we're saying is, given everything that we know right now, what's the best estimate of the risk? And that's the question we're trying to answer. We are not trying to say, hey, this is our process and this is how we rate this company. Instead, we're just saying, given everything that's available, what's the best we can do? And for a rating agency, I think this is still valuable. I think it would be especially for some of the bigger investors and bigger investments.

Rajiv:

I think some of the processes that the rating agencies have for due diligence and underwriting will still be very valuable, but I think solutions like Martini will be far more powerful for two big parts of the credit workflow. One, pre-trade due diligence, when you have hundreds of deals are coming in and you're trying to figure out which ones should you go after. That's one place where we think we'll have big impact. The other part that I'm frankly very excited about is portfolio monitoring and credit surveillance. Like once you have portfolios, you could just be a S&p 500 company doing business with maybe 10 000 companies and you're worried about your accounts, receivable risk and you need to monitor risk on a daily basis, and I think that's where a solution like martin you it's super powerful because it lets you do things like uh, how do you monitor your portfolio on a daily basis, how do you see what happens to your business in the face of inevitable big events happening? Or things like, how do you ensure your portfolio on a daily basis? So those are kind of big use cases that I think Martini unlocks, which is going beyond the world of just labeling companies for risk but saying how do I operate in this world?

Rajiv:

And in fact, a couple of years ago when I started, there used to be a slide I used to show all the time, which was? You remember that incident of the ship getting stuck in the swiss canal I think it's the ever given, or something like that and like the ship swerved just a little bit, got stuck in the sand and then all these ships got backed up behind it. And then price of oil next thing, you know, the price of oil was shooting up just because, like, the ship had drifted like a few feet and this is just like a fantastic example of the butterfly effect and I don't know how many companies got impacted then. And uh, you know, like you know, it would be difficult to understand what's happening. And what we are excited about now is like you can just type out those situations in our scenario builder, and it'll tell you what happens, which companies get impacted, and so it's just very exciting to be able to see that come to life and being able to have tools to operate in such a world.

Jim:

And what advice do you have for firms that are just beginning to explore AI-powered scenario modeling?

Rajiv:

The first thing is just getting familiar with the tools. The second thing is building a vocabulary. I feel like a vocabulary does not necessarily exist for understanding what kind of scenario should you worry about? Because I know that the regulators require banks to do some of this. They have stress tests and those stress tests are mostly around interest rate hikes or modeling like a past scenario saying hey, if the GFC happens again, what happens to your portfolio? Do you have enough capital to cover that?

Rajiv:

But I think scenarios can be so much broader, risks can be so much deeper, and just being able to build a set of scenarios and responses and action plans around each of those will just make firms much more resilient, and I think so. That would be my first piece of advice, which is just saying what does your playbook look like? What risks are you set up for, what can you handle, what can't you handle and which of these are consequential. And the rate at which these scenarios are unfolding now is mind-boggling. And like I was in New York a few weeks ago, the week of the tariffs, every single day the tariffs changed and like for the traders on the floor, I was talking to someone at a japanese bank and she just threw her hands up in the air saying, hey, I just don't know how to respond to any of this because, uh, things are changing way faster than I can, uh, uh, anticipate and or I can plan for. So, given that these, now these scenarios have just become, you know, on a daily or hourly basis, they're unfolding at that rate it's just so much more important to have a good handle on what kind of scenarios could come up.

Rajiv:

Two is what's your response plans? And three is what kind of levers do you actually have? Do you have enough levers? And I think that's the third part. The third part is where I think a lot of users will end up introspecting a lot, because what good is being able to understand scenarios if you can't act on it and just being able to build the levers to operate in this?

Jim:

that's the part they need to focus on so unfortunately we've made it to the last question of the podcast. We call it the trend drop. Uh, it's like a desert island question. And if you can only watch or track one trend in ai and investing, what would that be?

Rajiv:

so well, it's the obvious one and but it's uh obvious and very powerful it is the track of, like you know, agentic AI in helping making your decisioning and understanding what's happening and shaping your portfolio and your decisions. And the way I think of it is how can agentic AI be a co-pilot for you? So you're already seeing this in code, in coding. You have companies like Cursor who act like engineers for you. You can just sit and talk to it all day long and it generates phenomenal pieces of code that you can use to get. That does work for you.

Rajiv:

So what's the analogy in finance? So it's going to be an assistant which sits with you and helps you think through all these huge data sets, helps you think through all of the scenarios, helps you think through decisioning and where it will go from. There is, once you have the decisioning, the next thing is the actions, and so AgentDK are making actions for you, like deciding limits, deciding investment decisions and then eventually shaping policies. So then you could be an investment manager who's spending time thinking about what should my portfolio kind of roughly look like to meet the investment goals of my LPs and then working with your agent to craft that policy and execute on that policy. So I think that is becoming real much, much faster than people anticipate, and I would keep an eye open for that.

Jim:

Well, rajiv, I want to thank you so much for your time, your insights and the information you shared with all of us. Thank you so much.

Rajiv:

Thanks, Jim. All the questions are fantastic. I absolutely love being on here. Absolutely love being on here.

Jim:

Thanks so much for listening to today's episode and if you're enjoying Trading Tomorrow, navigating trends in capital markets, be sure to like, subscribe and share and we'll see you next time you.