We cannot guarantee the accuracy of this transcript.

Erika van der Merwe (EM): The pursuit of alternative alpha has been held as the next frontier in quant investing. Using artificial intelligence, algorithms and vast volumes of new data types to explore and tap sources of returns does sound intriguing, especially when times are uncertain and alpha is elusive. But, is this hype, false hope or real opportunity? And what does it mean for investors? We’re combining a quantitative and a fundamental perspective to answer these questions. Around the table with me, we have Weili Zhou, she is Head of Quant Equity Research at Robeco. Daniel Ernst, who is a Senior Investment Analyst at Robeco with a speciality in global trends. And Mike Chen, Head of Robeco’s Alternative Alpha Research. Wonderful having you here in town with us and around the table, welcome.

Daniel Ernst (DE): Thank you, Erika.

Weili Zhou (WZ): Thank you.

Mike Chen (MC): Thank you, happy to be here.

EM: Let’s get started with these terms. These are really big names, big terms being thrown about, so what precisely is meant by AI, machine learning, and alternative data, Mike?

MC: Yeah, so artificial intelligence, I think, is a term that’s actually used quite a lot, but it is actually quite distinct from machine learning. Roughly speaking, artificial intelligence is basically you can think of something as sentient. It can think for itself. It can actually have intuition and can infer from knowledge, from context, from its experience, and make judgment and decision on stuff it hasn’t seen before. That’s artificial intelligence, that’s basically like human intelligence, right? What we’re doing in finance, actually in general, is a subset of artificial intelligence called machine learning. Machine learning, rather than being able to make inference from experience and make estimates or decisions about things it hasn’t seen before, cannot do that. Machine learning is basically… What it does, at a very basic level, is: it maps inputs to outputs from results that it’s seen before. It can make linear mapping, but it can also make non-linear mapping. That’s what makes it so powerful. It can learn from a vast amount of data, and it can appear like it’s sentient, but it actually isn’t. It has to see the data before it knows what to do with it. If it hasn’t seen it, it can not make inferences or decisions.

EM: Then what about alternative data, the unstructured data we keep hearing about?

MC: If you think about it, traditionally in finance, historically the data has been very structured, very tabular. So these are data you get from market returns, you get from financial statements. These are data that can fit in a table, really. But obviously, we all know now that the world is awash in data. Not just these very structured, relatively small, compact data. You have all sorts of information that’s out there. The internet is a huge source of data. Your GPS, where you go, it gives off geolocation data. Your cellphone. Satellite images. Your credit card spending. That’s all data. Actually, what people say in both textual and audio waveform. That’s all data. Alternative data is really anything that’s beyond the traditional financial statement and market return type of data, and us investors are trying to harness a vast amount of unstructured data and to see if we can get further insights into our investment process and deliver alpha for our asset owners.

EM: Right, so we’re going to pick up on this and go far deeper. Daniel, it feels like a new paradigm that we’re entering. What are you seeing as an analyst of global consumer trends?

DE: Right, I think you already see AI, rather machine learning, fitting into your everyday life. As Mike mentioned, your navigation system is routing the best way to get home to avoid traffic. When you open an email and it starts to autocomplete what you might say in a response. You’re seeing it in these small ways. But we’re also seeing it, I think, in bigger ways. People looking at this for medical research, for drug discovery. But, I would highlight two trends that are already here today that are actually monetizing, or to put it the way Facebook’s first CTO in 2011 put it, ‘The brightest minds of my generation are attempting to help people click on ads.’ And so, AI, or machine learning, has become the tool of choice in creating content or filtering content in a way that you want to see it, and then placing an ad in front of that. So there’s a way that actually you’re seeing, and advertising is a 500-billion-dollar business globally, so you’re actually seeing it there. TikTok’s algorithm around how it… It’s kind of eerily good at deciding what you might want to see next, but it also places an ad in that way. Then the other one you hear people talk about a lot is autonomous vehicles. But, long before you get to autonomous vehicles, you actually can have these things notice that there’s a car stopping in front of you and hit the brakes before you. While there’s been some notable accidents with autonomous cars from Tesla and others, what the data shows is: when those features are engaged, there are 90% less accidents. And then the other one, where autonomous is actually not about vehicles, but it could be in controlling systems. So, Google applied their DeepMind division to their data centers, and in that process they discovered they could save 40% of their energy costs by altering the cooling systems in the data centers. So you see it in these tiny small ways, in your email, and you’re seeing it in the news on autonomous cars, and then it’s helping to green the planet by turning off the lights when you don’t need them on.

EM: What I’m hearing from you, Daniel, is: it’s everywhere, it’s also generating so much data for us to use, and it’s also helping us in many ways. How is it helping us within asset management? And what’s the objective? What’s the purpose of machine learning and alternative data here?

MC: In asset management, the goal is always the same: deliver investment results and solutions for our clients. And I think alternative data, machine learning, alternative alpha just really opens a whole new frontier. Again, traditionally, from a quantitative perspective, we’ve been limited to relatively small sets of data, or tools have been relatively constrained as well. We’re mostly limited to discovering linear relationships. But now, with this vast amount of data, and these powerful algorithms that can discover non-linear interactive relationships, and also, how did people talk, their sentiment, what they’re inferring. Are they actually being evasive when they answer questions? These give us the capability to answer investment questions that were always interesting, but some of which until recently we had no tools and data to answer and to test if our investment hypotheses are correct. So, I think that just really opens up a whole new frontier of research for us investors.

WZ: I think what’s adding here is the explosion of computational power we have seen in the last decades, because machine learning, AI, is not a new thing. Back in the 1970s-80s actually there have been machine-learning models, but the use of cloud computing powers and all these that brought the research and quant investing to a much higher level, that you can really dig much deeper to find meaningful relationships, non-linear relationships, that you can put all those alternative huge data sets into your proposal and extract very important hidden information out of it. I think, never before I’ve seen the financial and investment industry being so linked to technology nowadays, and this is also quite a change compared to 20 years ago, for example.

EM: Right. If I were to ask you, why now? Why is it now such a hot topic? I think you’ve answered the question. We’ve got the technological capability and we have more data, but it does seem to have become a hot topic. So why is it such a hot topic right now? Just to give you some color for this question, here’s a quote from Marcos Lopez de Prado, he’s Head of Quantitative R&D at ADIA. He was speaking on The Parkview Podcast.

Audio fragment: Traditional data sets have very little alpha left. Why? Because they’re relatively easy to model, they’re accessible, they’re standardized. And as a result, it is the trodden path. Everybody has access to them and can try and develop a strategy. So, what they call macroscopic alpha is distinguished, is gone. But microscopic alpha is more abundant than ever. Why? Because on the one hand, you have data sets that were not available before and that are very difficult to manipulate, they’re very difficult to model. So that gives you access to opportunities that were not accessible before. So that’s on one hand. And on the other hand, you have the errors made in strategies and the quants that still utilize the old techniques and those errors can be exploited by machine learners. So, as a result, microscopic alpha is more abundant than ever.

EM: Any comments on that?

WZ: I think Marcos is exactly right. I think part of the reason why it’s been harder, the traditional factor, is because how successful us practitioners and academics have been in explaining sources of return. So you think about CAPE, you think about Fama French, these factor investing types. They were major discoveries, but unfortunately they’re relatively easy to replicate. It wasn’t easy before it was discovered, but now, since everybody knows it, it’s easier to replicate. So then, you always… The question is: what’s next? Whether it’s fundamental investing, whether it’s quant investing, whether it’s investing 200 years ago or investing today, one of the fundamental laws of investing is that even ideas can monetize. If too many people invest in the same way, you can guarantee that there is no return. So since these more traditional factors have been relatively well-known now, there is a very natural impetus for us in the investment industry, trying to see if we could be somewhat differentiated, trying to still seek alpha, which is always desirable, no matter what the economic or environmental or technological development is. It just happens so that, because of the availability of data, because of the availability of advanced algorithms, coupled with easily accessible massive computation power, we now have the capability to go for this microscopic alpha that Marcos talked about.

EM: So, take us into your rooms where you work and where you think, at your desks, what is Robeco’s approach here to machine learning, to alternative data? What are you up to?

WZ: I think for alternative data we are skeptically optimistic. There’re so many offerings on the street, and nowadays we’re talking about 2,000 vendors, and everybody’s trying to sell very specific niche information to us. It could be satellite pictures of the parking lot, it could be a very specific industry, in the local Chinese or Japanese market. So, there are abundant choices out there. And as a company, we have limited resources, we have limited hands, so we have to do the pre-screening and think very carefully what are the criteria we want to adhere to while making the choice. And also making the choice on starting something, not only on the acceptance end, but also on starting a project in the first place. So I would say that we would first of all stick to economic rationale. I think that’s also the place where we profit so much from a firm being strong, both in quant and fundamental. For example, if we see some credit data information or we see something very specifically related to a sector, for example healthcare, I think we have so many expert fundamental analyst colleagues to approach it and say, “Do you think this data set makes sense? Would you as a human make use of it in the first place?” And that sanity check for us I think is crucial in already filtering out maybe 50% of the vendors that just brag about their quality and the uniqueness. And once we start to research on alternative data, for example, we also have a couple of rules. First of all, the permanency risk. Will the data still be there after a year? Is it legal, for example, to scrape in this country? Will this data provider still be around in two years’ time? I think this is crucial, because a financial statement, any report, that’s something coming up every year, every quarter, but alternative data, there is no guarantee.

EM: It’s clear from what you’ve all described that this is indeed a very complex new field. What does it take for you as an asset manager and investor? What infrastructure do you need to be able to do this right?

WZ: I’m super happy you are asking this question, because you are asking a system question, right? This is typically not visible to the outside world. People talk about models and portfolio, but system…

EM: It’s abstract, right?

WZ: Exactly. This is below the water, but it’s so determinant. Determining your capability of whether you can breed the next generation quants in the coming ten years. This is also the field therefore in which we have invested very heavily in the past three years. For example, we’re now operating on an infra platform that is so well-connected to, for example, cloud computation, so well-connected to different data scout centers for alternative ideas, and also very well-connected to AI algorithms from different tech firms. For example, the latest discussion we had is to look into the Google algorithm and to see how it can be applied to our NLP analysis. Also, it’s very much connected to the open-source comprehensive machine-learning training platform. All these actually are very powerful, and with these infra-developments in place, I think we’re so equipped to on the one hand push for the new frontier of factor investing, incorporating all the new techniques. On the other hand, to explore for alternative alpha and bring new standalone strategies to the offering of the whole quant house.

MC: Actually, I would just add that… Machine learning, actually modern research, really, it’s really a multidisciplinary endeavor. You need to have infrastructure, which is usually not exciting, usually you don’t talk about it, but it’s critically important, it’s like the foundation of a house, you need to have a good foundation of a house. You need to have people with domain knowledge, knowing what they’re doing. As you are aware, machine learning is very powerful, but it’s also very easily misused, so you need to have people asking reasonable questions, economically-based, fundamentally sensible questions. You need them to have technical competency. We don’t need to invent the next greatest machine-learning algorithm. That’s not really our job. But we need to be able to understand the strength and capability of each algorithm and how they may be used in a financial context. That’s really important. I think I would probably call this part of the infrastructure or just part of the ingredient that it takes to make machine-learning research work in general is: you need to really have a good culture. The culture comes up a lot in this podcast, but I think culture is vastly important, especially for a research organization. You need people to be very open-minded, you need people to be very direct and very honest. You cannot have a very hierarchical setting where if the senior researcher says something the junior researchers are perhaps intimidated or afraid to voice his or her opinion. A good idea can come from anywhere. Nobody has a monopoly on good ideas. So we want to be able to foster that. Speaking as a non-Dutch, I really see that the Dutch culture, which is very flat, very direct, actually is a huge plus when it comes to research. I really find that to really be kind of a hidden edge, actually.

EM: You said that because many of your colleagues at Robeco, with the Rotterdam head office, are Dutch.

MC: They are Dutch, yes. But, we actually have a pretty diverse team. As you can see, the two of us are actually not Dutch. We do have quite a bit of Dutch, but we have a very diverse team. I think what’s really essential is that we have a lot of the Dutch culture, where it’s very open. Everybody can voice their opinion and nobody is really intimidated because some senior person said something, so I better not contradict. We’re all very evidence-based, and I think that’s really quite important. Like I said, I think it’s a hidden edge that this research team has.

EM: Maybe, Weili, a question for you: how does your approach differ from what you are aware of is happening elsewhere in the industry? It sounds like what you described, with architecture, the infrastructure, that there’s also a degree of first-mover advantage here. So, perhaps you’re quite early on in this phase.

WZ: I would say infer is definitely an edge for us. I think that’s also something positively surprising to Mike, who joined us lately. Apart from that, we actually are still very rigorous with testing and also acceptance of any new innovative ideas. It’s not that just because it’s new, it’s very novel, and the bar is lower, we’re not saying that, “For whatever evidence or selective bias that we see there, we would accept a model or signal into the process just because commercially it’s very attractive, that you have this fancy new signal.” I think we really, like I just mentioned early on, have our set of criteria and we say that additivity, permanency risk, trading cost profitability, after-cost trading profitability, all these should work before we give a green light to something new or something innovative to get into our process. I think also, the speciality of Robeco is the joint effort from quant and fundamental colleagues. We have a lot of close conversations on what we could do together.

EM: Yes. Another speciality at Robeco is sustainable investing. So how does this work along in that field?

MC: Great question, actually. If you know a few things about sustainable investment, maybe you know that sustainable data is hugely lacking. They typically have a lot of gaps. A lot of it are estimated. A lot comes with huge time delay, in low frequency. So none of these conditions are great for when it comes to investing, because you want data that’s rich, that’s timely, that’s very frequently updated, not estimated actual data. So, sustainable investment is hugely challenged from a data front. A second difficulty with sustainable investment is that a lot of what it’s trying to measure is very intangible. Corporate culture is a sustainable investment. Various stakeholders’ sentiment are a sustainable investment. Whether a company is ethical or not is something that is very difficult to quantify, using traditional means. So, all of these alternative approaches, data, machine learning, algorithms, etcetera, can actually come in and help you measure various aspects of it. So, sentiment, obviously you’ve seen natural language processing, you can measure various stakeholders’ sentiment. Whether a company behaves ethically or not, you can actually see if they have a lot of negative news coming on from news organizations, from NGOs, from government fines, just from, “Do people talk badly or well about a given company on the internet?” So, I think the intersection of sustainable investment and alternative alpha or the alternative approaches is actually a very, very powerful intersection. It is one of the few frontiers, I really view, truly relatively underexplored frontiers of investing, and I think this is just a hugely exciting area. As we all know, I think there’s an awareness that it’s not good enough just to actually make profit. Listen, we live on this earth and the resources are not infinite, we have to be able to get along with each other, so you have to figure out a way to do it profitably, obviously, because not being profitable is also not sustainable, but you need to be able to consider everything, and I think all these alternative techniques bring so much to the table when it comes to sustainable investing.

DE: I think, just to chime in for the human race here. I think this is an area where…

EM: Says the fundamental researcher!

DE: This is an area where fundamental investing helps, because as Mike points out, the ESG data is laggy, it has a lot of gaps, and so actually having… Robeco has a very large sustainable investing group. They’re looking at companies and measuring it and studying it on their own, separate from the radiancies and the data that comes in. I’ll give you an example: we had a company in our portfolio whose product mission really reduced greenhouse gasses, but it had a really low, almost I think a negative ESG score, because the metrics they were looking at weren’t looking at the product impact; they were looking at more traditional operational elements to the company, which were probably accurate. But, they weren’t measuring the product impact, and so our own SI team was able to point that out to us and say, “It’s in our portfolio, with a low ranking, which impacts the way you might look at it from the outside in portfolio holdings.” But the actual product impact was very positive and you don’t see that. I think in an area where the data is imperfect, and I think you have that a lot in structural change. I think the shift to a greener economy, the shift to new forms of energy is one where the data maybe wouldn’t suggest that it’s profitable. You look at data on, people still have these pre-conceived notions of renewable energy being lossmaking, but today the actual results are: it has a lower cost of production than traditional metrics. But the lag between when that shows up in the market and when it’s just in profits I think doesn’t show up in the ESG data.

EM: Right. Listening to you on this, it’s very clear that you’re excited about it, putting a lot of resources, thinking and planning into this, but I have to challenge you, because in preparing for this I read a lot, watched videos, listened to podcasts. Is it realistic? Do we know for sure that machines can learn to invest and indeed create these new sources of data and correlated data, that one third, Mike, that you said, was still left on the table? Here is an academic paper entitled “Can machines ‘learn’ finance?”, published in the Journal of Investment Management, 2020. I know for sure one of the authors works for a quant firm, so he is not in some theoretical la-la land. He says, “The industry’s collective machine-learning marketing hype must be tempered. The gains are evolutionary and not revolutionary. Asset management and return prediction in particular is a small data science with low signal-to-noise ratios, making it very different from disciplines where machine learning has thrived. As a result, adapting machine learning in finance is a more difficult proposition than many commentators appreciate.” This is someone working in theory and practice highlighting this. How much hype is there here?

MC: Great question. I think I know who wrote that article. There’s certainly a huge amount of hype. Because it’s sexy to talk about. I used to work at Google and there used to be a joke at Google while I was there, that we spend billions of dollars, hire the smartest people in the world, thousands of them, to identify cat pictures. Cat pictures ten years ago was a very difficult problem. It’s difficult, but it’s easy in the sense that while you look at a picture of a cat last year, you look at a picture of a cat last month and you look at a picture of a cat today, it’s not going to change. Within the timeframe we’re talking about, a cat is a cat is a cat. But, guess what? The market is adaptive. The market is… When people start reacting in another way, the market prices it in. In a sense, the market itself is kind of like machine learning. It learns what people… Because market participants learn. So, it's adaptive, meaning that the way it behaves changes over time. So, this is one huge difference. I agree with you, I think machine learning is somewhat overhyped, but I believe machine learning is very useful. It cannot completely do everything, so there’re still roles for people. But, it is a very powerful tool and it is quite evolutionary in a sense that it expands our capability and our toolsets. Before we could only look at linear relationships. Now we can profitably look at non-linear interactive relationships. But does it answer everything, or even a majority of the questions you might have as an investor and you should think about? No, it doesn’t.

WZ: If I may add, actually, I think it’s an excellent point that the market is indeed adaptive. For example, the earning call transcripts that we ask a machine to analyze, and then of course you start with a bag of words: positive, extremely, great outlook. These kinds of things a machine learns. But guess what? All the executives also learn these bags of words. And then you see that these executives actually, knowing that the machine is at the backend analyzing their speeches, are changing their wording and changing their phrasing, and somehow the machine couldn’t pick up the nuances anymore, and actually, in such a way, the machine needs to adapt as well to dig deeper and to be retrained by the fundamental investor, by the quant investor, and say, “Now it is the new bag of words that describes the positivity and the negativity.” And I think that was also a very interesting experiment we had with our fundamental team two years ago. We asked men and a machine to do a competition. So we just gave the fundamental analysts I think the ten best recommendations and ten worst recommendations from a machine and asked them to rate these reports. And guess what? I think it’s very interesting. On the top recommendation from the machine, analysts tend to agree, but when it comes to the negative rating, where the machine was very negative, but analysts just see a different interpretation and they’re kind of neutral to these earning call transcripts. This is where we see a huge divergence and we need to have these human colleagues as the trainer and tuner of the machine and making sure that it is doing what we would want them to do and also making them adapting to the world that is evolving around it.

DE: I think that’s a great point of how the executives learn that bag of words and they were able to script their calls better. But, I think for a long time, many people in fundamental research have sort of found the earnings call to be repetitive and not providing signals, so nothing about that question of a signal from a computational standpoint, but just from a “Did I learn anything new?” standpoint. Then you find that the same CEO speaks on a podcast three months later on something totally different, but he makes some comments about his worldview or where he sees the markets going, or some hobby that he has and, “That’s interesting.” You pick up something with a strong signal in a non-structured environment that the algorithm maybe would miss, because it wasn’t in the earnings call.

EM: So, the human oversight remains important, Daniel. We need you.

DE: I have to keep making the plug.

EM: Indeed! So, what would your advice be to anyone listening with their own investment firm or even an asset owner wanting to go this route? What advice would you give them? Because there must be some learnings, but also some warnings on what the constraints are.

MC: I think rule 101 of machine learning is: it is still a very rapidly evolving field. But the threshold to entry is lower, although you still need quite a few elements, like domain knowledge, technical competency, data, but the threshold is lower, so I think…

EM: But not for a boutique asset manager, right? A small manager with limited assets under management, could they afford it?

MC: They could do it at a limited scale. I think the number one recommendation I would have is just to experiment, pick it up, try it. I think it’s really… You need practice. You need to gain experience. You need to find… I would say there is no one-size-fits-all way to do machine learning. Obviously, you need to be able to program. You need to understand some basic linear algebra. You need to have access to some computing power. But every firm’s setup is different. Obviously, being a relatively decent-sized firm such as ourselves, with management support, that we’re able to build up a very robust and scalable infrastructure, is a huge edge, especially when it comes to maintaining and very quickly iterating. But even a small firm can iterate, can learn. The key is: get your hands dirty. I think that’s really number one.

WZ: And never be afraid to be late in the game. I think there is first-mover advantage, but sometimes there is also second and third-mover advantage. So, you just leverage on what people have found and also leverage on what is already developed out there. So, for NLP study, for example, if you started ten years ago, it’s only a bag of words you can make use of, it’s a dictionary. But nowadays, look at what Google could offer and OpenAI, these kinds of platforms could empower you with highly-advanced, human-like algorithms. That’s a totally different ballgame. First of all, I think, don’t be afraid to start late. And secondly, think carefully on how to buy and build. So, something you can buy, something you can build, but with a smaller or mid-sized company maybe you can buy a bit more than build, because you can also say, “I start with a light investment via some, like you mentioned, Mike, experimental platform and start with some very specific cases, like I want to try it with one or two capabilities, if it’s in the data-driven process, how would it look like? Don’t do it like overall everywhere. If you have 20 products, don’t do it with 20. Maybe select one or two to experiment and then you get a very tangible feel on how it looks and how that compares to your own experience, and then to see whether it’s smart to roll out.

EM: Let’s draw it all to a close now. What do you believe the future holds? Some CFA research said, at that time, that was 2019, it could be quite different now, at that at stage, only 10% of respondents in that survey had used ML techniques over the previous 12 months. What do you expect to happen over the next five years?

MC: I think the evolution continues. It’s not a revolution, but an evolution. So it’s going to become more commonplace, more people are going to adopt it. Just like quant investing 20 years ago or 30 years ago, when Robeco started quant investing, it was probably one of the very few houses on the street in the world that did it, in the 90s. Now it’s more common. People evolve their technique, become more advanced and everybody starts doing slightly different stuff. I think it’s really going to be the same way with machine learning. People are going to adopt; people are going to come up with different insights and different ways of applying and then it’s going to… Maybe the main ideas are the same, but the changes on the margin will cause enough of a differentiation, that you might still be able to extract alpha. I think it’s quite exciting. I don’t know when, if or when fully automated machine learning comes into place, but if Google can build machines that can beat the Go champion of the world, this is perhaps not beyond the realm of possibility. I don’t know when that’s going to be, but I think if that happens it could be very exciting.

EM: Final remarks from you, Daniel?

DE: I think, listening to Mike talk and mention evolution, I’m reminded of a story that Steve Jobs liked to tell, where in a study in the Journal of Nature, looking at the efficiency of motion of different creatures, so how many Watts would you need to propel yourself. A condor or a cheetah, was the most efficient and humans were kind of in the middle of how much energy it would take you to move forward, say a meter or two. But then someone had the idea to put a man on a bicycle, and that ended up being the most efficient by far. We’re sitting here in the Netherlands, so it’s also…

EM: We’re back to the Dutch!

DE: The bike lanes are really helpful in this regard. But, he concluded that the computer for him was a bicycle for the mind. So, this more efficient way of doing things, in the same way that the bicycle can propel a human faster, the computer can help propel human intelligence. So, I think in that evolutionary way, AI will do the same thing for human learning.

EM: And Weili?

WZ: I agree with the comments from Mike and Daniel. I think it’s a facilitator, it is a bicycle for our great mind, and at least we expect them to unload and unburden researchers from the massive work and make us a lot more productive and we can do more the thinking-designing work than the repetitive, for example, backtests or reporting work. I think that’s where definitely we see it adding the value.

EM: Weili, thank you very much. Daniel, thank you. Mike, thanks to you. It was great listening to your power-packed perspectives on this topic.

WZ: Thank you, Erika.

DE: Thank you.

MC: Thank you, Erika.

EM: And thank you for listening for us. It was great having you part of our conversation. We’d love to hear from you, so please let us know if you have any feedback or suggestions. You can contact us on podcast@robeco.com. You’ll find all of our podcasts on your favorite podcast platform as well at robeco.com.

Male voice: Thanks for joining this Robeco podcast. Please tune in next time as well. Important information: this publication is intended for professional investors. The podcast was brought to you by Robeco and in the US by Robeco Institutional Asset Management US Inc., a Delaware Corporation, as well as an investment advisor registered with the US Securities and Exchange Commission. Robeco Institutional Asset Management US is a wholly owned subsidiary of ORIX Corporation Europe N.V., a Dutch investment management firm located in Rotterdam, the Netherlands. Robeco Institutional Asset Management B.V. has a license as manager of UCITS and AIFs for the Netherlands Authority for the Financial Markets in Amsterdam.