We do not guarantee the accuracy of this transcript.

This podcast is for professional investors only.

Erika van der Merwe (EM): The buzz around artificial intelligence has reached fever pitch with growing awareness of what this technology can do. But what does it all mean for investors? Can the power of AI be harnessed to create better investment outcomes?

Welcome to a new episode of the Robeco podcast.

EM: We've invited three specialists in this area to discuss the various possibilities and how Robeco is approaching what it calls ‘next-generation quant investing’. My guests are Weili Zhao, Head of Quant Equity at Robeco, Wilma de Groot, Head of Core Quant Equity, and Iman Honarvar [Gheysary], Director of Quant Equity Research. Welcome, Weili, starting with you. Where are we in this evolution of quant investing?

Weili Zhao (WZ): I think to have an understanding of where we are today, we need to look back first. If we look at the history, it's fair to say that technological advances have always played a deterministic role in terms of bringing changes to how we work, how we live. So about 100 years ago, television and telephones got introduced, and since then news started to travel very fast and investment decisions could be made more timely and more accurately. So that's a game changer.

If you look at 30 years ago, the introduction of personal computer terminals to the daily work floor also changed the way of work, and suddenly the processing of data and financial reports became a lot easier. And that actually also opened the door to quant investing, namely systematic investment strategies that got tested and managed based on models and algorithms.

So if we look at today [in terms of] technological development, I think the progress in the last decade was stunning and particularly in three aspects. First of all, I think we've seen the explosion of data. Every two years nowadays I think, the total amount of data that was created [in those two years] is the sum of all the data that has been generated previously in human history. Everybody is generating a lot of data every second – your mobile phone, your GPS. So when we look at a company, for example, looking at the investment case, it's not only the stock price, it's not only the financial statement. It's real time information around a company. What did the CEO say in the news? What did the employee complain about on websites? What did the competitor bring to the market? What are customers liking on social media? All this is available information being so rich and that is just staggering.

I think the second change we see is the explosion of computational power, and that's another game changer. According to the law of accelerating return, which is also known as Moore's Law. So the number of transistors on a microchip, it doubles approximately every two years. So if we look at the curve, it is exponential growth in computational power. In the last 50 years we kind of managed to have that pace. And that's realizing Moore's Law. And the last five years is also even more geared up in combination with cloud technology. So in theory, actually, we have almost unlimited access to computational power, which allows us to dig deeper, to work on more sophisticated machine learning models, to understand the interaction between all the signals, hundreds of them in the pool, and also enable us to develop more humanlike algorithms.

And the third one, I think it is all very fascinating to development of AI technology in industry. So I think we are lucky as a quant investor that we could stand on the shoulders of giants, looking at the leading tech firms like Google, OpenAI. They've made progress and they've been pushing the frontier of AI technology and algorithm. They've been astonishing us with large language models so that from our quant sides, we could quickly adapt and translate their latest developments into our strategy and see whether it can be deployed instantly or not.

EM: Robeco itself has a very long track record in quant investing. You yourself, you personally had a key role in this. Where's Robeco in this evolution?

Wilma de Groot (WG): So since 2004, we run our quant equity strategies live. And it started with enhanced indexing strategies in developed markets. Later on we also extended this to emerging markets and also to more active strategies. Examples of that are conservative equity strategies or active low risk strategy, and also later on, factor investing strategies.

And if you look at the stock selection models that we use and that we exploit, you see that we use many well-known factors such as value, quality, momentum for return purposes, but also low risk for conservative equity strategies. And a lot of research that we've done was actually to enhance these type of factors, mainly to reduce risk and to achieve better risk-adjusted returns. For example, we removed style biases in our momentum factor to reduce drawdowns during reversal periods. And we also reduce distress risk related to our value factor to also achieve better risk-adjusted returns.

But it's good to know that in addition to these proven factors for which there has been a lot of academic evidence, we also explored innovative signals based on alternative data sets in all our quant equity strategies. And the main reason to do that is for diversified alpha purposes. We observe that these type of signals have a low correlation in general with these more well-known factors, and that helps to reduce drawdowns and also to improve our information ratios.

EM: It's clear just listening to you that it's just a process of ongoing innovation. You're constantly on your toes, just bringing everything in.

WG: Exactly, exactly.

EM: So we heard Wilma using this term ‘alternative data sets’ and really also talking about data, computational power tooling, all of this happening. So let's start with the alternative data sets. What exactly are they and how can investors use them to generate better returns?

Iman Honarvar (IH): Very good question. Actually, like Weili and Wilma mentioned, we are generally surrounded by data. We are living with data. Even right now that we are speaking, we are producing some data. And this data has been helping us in different fields, from healthcare to making life-saving medicines to finance and transportation. On a weekly basis, we are in contact with a lot of data providers. We get a lot of data, we try them. Quite often, we find they are not very suitable for the type of investment strategy that we have, but some of them are really helpful and help us to diversify and enhance our portfolios.

I can give you some examples. For example, maybe two decades ago, three decades ago, Robeco was one of the pioneers in ESG investing. But 20 years ago, 30 years ago, even for the pioneer, ESG investing was very different. It was ESG was mostly based on some companies surveys. Companies would say how they treat, for example, disposing the waste. And then according to that, we could score them positively or negative from environmental perspective, how they treat their employees and so on and so forth.

As you can see, this is quite crude. And this was the cutting edge of 20, 30 years ago. Nowadays we have much more data. We are surrounded, for example, by text. There is a lot of text about the earnings call of the companies. The report and analysts are writing the speech that the CEO is giving about the company, the news that any minute can come up about companies.

This text is a very fertile source of information for us. For instance, we have been looking at these huge amounts of text, and if we analyze them using natural language processing, we see we can easily score companies in different dimensions. We can give basically find the company profile from different dimensions. For instance, if employees are very satisfied about their company and they write a very positive review of their company as a good employer, then just from the anonymous review of the employees, we can find their score from a social perspective.

Or for example, if they talk very positively about the innovativeness of their employer, we can see which companies are more innovative and versus the other ones, or from a fraud perspective and so on and so forth. So I would say the computational power that we have, the data that we have has opened the gate for a lot more to come going forward.

EM: Well, all well and good. Very exciting to hear about the data, but is it really usable given that – I'm sure lots of this data is pretty new. You don't have a long time series.

WG: Yeah, that's true. And I think one could argue that the data history or the data coverage is relatively low, and that you should therefore not use it in your strategies. We acknowledge that the data history for several of these data sets is indeed quite short, but long enough to do decent research, and so also to consider adding these type of signals in the strategies.

And we think it's also important to already start using it because we expect that the –we can already we know that the data coverage will only increase. And that by already starting looking at this data, you also learn more from it. And you get only better and higher quality data.

And to give a nice example of this is when you look at emerging market data. So in around 2000 we developed our emerging market stock selection models. And back then the data history of emerging markets data and also the quality was of course much lower than compared to developed markets. And we tested the same type of factors that we also found that worked very effectively in developed markets. And despite the short history we found these were actually very effective. And by building these models, we also learned more about the data and we improved our data quality.

Actually, the data providers that we made use of, we actually teach them how to improve the data and correct errors. So in the end the whole investment industry profited from that. So I think this is a nice example. And also to allude to your point on sustainability, Iman. I think this is also a nice example of one of the earlier, more alternative data sets that we used in the area of sustainability.

So we actually investigated around 2009/2010, ESG proprietary data set of [Robeco]SAM based in Zurich. This is very interesting information because we looked at the ESG performance of companies that participate in these questionnaires, and they have to provide a lot of evidence. By doing that, you get much more information than just publicly available on websites or annual reports. We found that companies with a high ESG score outperform those with a lower score, despite the relatively short history of like a decade of data. But also we found the same effect afterwards. So this is also an example of – you get more familiar with it. And also nowadays we actually have extended this type of sustainability data with other and more alternative data sets as well.

EM: Wow. Fascinating. So that's a bit on the data. And I'm sure you've written many chapters on this topic. Moving to the technology: machine learning. Let's take that as an example. What are the benefits but also the pitfalls of using machine learning for investing?

IH: Very good question. Machine learning is not something new. It was basically established half a century ago. It has been used in various areas. For example, in playing chess about two decades ago, it could defeat even grandmasters in playing chess or in facial recognition that was made about a decade ago. Finance is maybe one of the last ones to pick it up, but also for a good reason, because in chess and for example, facial recognition, the rules are more set.

In chess, for example, the pawn can only go forward or in facial recognition, the nose is always in between the two eyes. So there is more structure there. In finance, on the other hand, people can react differently to the same policy of the central bank. The world is more dynamic in the world of finance, and therefore this is one reason that a human should always be beside the machine. So human plus machine rather than human versus machine.
One other reason to mention is the data availability. For example, for chess you can put two computers that constantly play together and they learn from each other. But in the world of finance, we have, let's say, one century of reliable data sources. And within that one century we have maybe 20 economic cycles. So really, since machine learning models are data-hungry, we need to really provided a lot of data. And that's why the finance is gradually picking it up with more human help. Having said that, machines can quite often uncover patterns that are really hidden in the first sight from humans. And then it's the job of the economist to pick it up and bring it further.

So, for example, when you are investing in the short horizon, one of the signals that very often comes up is the short term price reversal, which basically means when the price of the stock falls over a short period of time, it tends to revert back because the price pressure has gone and then price comes back. But while training a machine learning to predict stock returns in a short period of time, we saw that actually the machine discovered some patterns that we couldn't see before.

For instance, it showed us that when the price of a stock falls for a distressed company, this doesn't revert back because distressed companies are in distress, and they are probably reacting to some bad news and it tends to keep going further down and down and down. So there are some hidden patterns that machines can uncover for the human and humans will give it economic color, and then they can bring it on further to production. There are many more examples that we can discuss about this.

EM: Let's take GPT as an example. As I understand it, my very limited understanding of GPT is that it is a subset of machine learning, and I think that's really come to the fore this year with ChatGPT. Now we all know about this. Does it – GPT – have applications in finance?

IH: Indeed. Interesting ChatGPT or GPT technology in general is, I think, one of the most revolutionary, or one of the ten most revolutionary innovations of humans over the past few decades, I would say, besides the internet and cell phones.

For example, the word Biden and President are very probable to happen often together. And therefore, if you ask ChatGPT, what is the job of Joe Biden? It says he is the president of the United States. It is very unlikely that the word Biden and teacher will happen together. That's why you don't see this sentence. But often it might also happen that because of some noise in the data or some reason, it can give you wrong answers, something that they call ‘artificial hallucination’ rather than ‘artificial intelligence’.

So again, this is an indication that humans should always be on board and walk together with the machine. Having said that, we are already using ChatGPT in different tasks and I think as time goes by there will be more things coming up with ChatGPT.

A few more examples – maybe I can give. I personally use it for summarizing papers and one other example, it’s actually a very good research assistant and a sparring partner.

EM: What about the other technologies – Wilma mentioned in NLP? Are you using that? What are the opportunities there?

IH: NLP is actually kind of the father of ChatGPT. ChatGPT is a particular type of natural language processing. And again, it's one of the fastest growing fields in machine learning. We have been using NLP for more than five years in this company. We have been using news sentiment. You know, every minute news can come up about companies and it can have a positive or negative charge. Basically, there is a machine learning model that is trained on the sentiment that this news can have, and therefore we can already constantly give a news score of positive or negative to news.

EM: So it's very clear amongst the three of you and I think [amongst] your teams that you are doing your research on this, looking at this, but is it already investable? Are you applying this technology and this data across your strategies yet?

WG: Yeah, we do. And well, we touched upon this example of sustainability as an example of an alternative data set. But maybe it's also nice to mention that we have built a separate kind of short term driven stock selection model. Actually the first draft – we developed that back in the days in 2009 and further developed of course, through time.

And this model is based on newer alternative data sets, but also newer techniques like machine learning techniques, also Iman what you elaborated on like discovering the non-linear relationships between signals like the reversal signal, you mentioned the future stock returns. So these are all very nice examples. Often, you see that these type of signals are more faster moving in nature. That's why we also use this as a kind of timing indicator to determine when to buy and to sell the stocks. We use it actually in all our quant equity strategies.

It's also nice to mention that the weight of these signals have increased through time. So while we started with a relatively low weight in 2010, the weight has increased. And the main reason for that is that, what's very important with these signals is because of this fast moving nature, the turnover of these signals is also much higher. So despite that you can add to your performance by combining these signals with more well-known factors, you also see that the turnover increases. So it's very important to be cautious about that.

Therefore, we've also spent quite some research on reducing trading costs. We use our proprietary trading cost model for that. By doing that we do not only lower the trading costs directly, but we also have the opportunity then to give more weight to these new, more innovative factors. Besides the use in all our quant equity strategies, we also launched early this year a strategy based purely on this type of signal. That's called the quantum strategy.

EM: Really applying it in practice. Wilma, you referred to some of the innovative work you've done in sustainable investing. Just talk us through some examples.

WG: Yeah, that's a good one. We have actually a couple of examples here to mention. And one is related to resource efficiency. And if you think about the airline industry that's maybe most easy to give us an example, is that if you look at all the different airlines, they use different kinds of planes. Some planes are more fuel efficient than other planes. And you can imagine this as the ones that are more cost- efficient and they have a competitive advantage compared to their peers.

So what we did is that we investigated whether the resource-efficient firms – and that goes beyond airlines, obviously – whether they also do better in predicting stock returns than the ones that are less resource-efficient. And we measured that by looking at the carbon emissions of firms as a proxy for the resource input. And we look at that compared to the sales, and we find the ones that have low carbon emissions compared to the sales. We see that these indeed have a competitive advantage. And that also leads to higher stock returns. And not only that, the nice side effect, of course, is that this also helps in reducing the carbon emissions of your strategy. So I think this is a nice example on the environmental side.

Also on the social side we have a nice example. Iman, you also elaborated a bit upon that: by using data that is related to employee satisfaction. And here the idea is that if the workforce of a company is very happy, if you have happy employees, they are more motivated and they are also more efficient. We first investigated whether they are indeed more efficient by using a measure of the sales – actually, you could say the output of a firm – compared to their labor cost basis. And we indeed find that firms with more happy employees are more efficient, they have higher sales compared to these costs. Next to that, we also see that this has predictive power for future stock returns. And this is of course nice because besides these additional returns we also contribute here on the social dimension of ESG. So this is another example.

And besides using sustainability for alpha purposes, we also use it for risk purposes. An example in this area is an innovation of a couple of years ago linked to climate risk in value portfolios. We observed that value portfolios in general, they have higher carbon emissions because value stocks are typically asset-heavy stocks or stocks in asset-heavy sectors. That leads also to generally higher carbon emissions than the traditional benchmark. So we found an elegant way to reduce these carbon emissions while not giving up on the return premium related to a value strategy. Also in recent periods, we have also seen years where energy prices went up. We have seen that the profitability or the returns of such a strategy is very much in line with the traditional value strategy. So I think this is another nice example.

EM: Weili, It's really clear just listening to how much focus there is at Robeco on next generation quant investing, but also quant investing in general. What does it take to stay on the cutting edge because there's so much happening. I mean, I'm just thinking of the mindset that it takes – the tooling, the architecture that supports it all.

WZ: There is a very famous quote I personally like a lot, and it says ‘culture eats strategy for breakfast’. So no matter how great the strategy ideas are on paper, you do need a group of talented people to execute it in the right way. If we look at next-gen innovation, next-gen quant investing, it's absolutely a team sport. And to do it well, we need a very strong and good culture. And that takes years or even decades actually to cultivate. In the Robeco quant team, I would say that we have a very strong culture to reward high quality research. So meritocracy really lies at the center of all the other beliefs. That's the competition of the best idea: no matter who proposes it. Let's say if it's a junior who has an idea and says “Weili, you're wrong. I have a great idea.” And he should or she should feel very safe to address it, and also to speak up.

Also, to make it work, you need to have a very flat structure. You have to get the direct communication to each other. And also you need to be able to embrace outside-in perspectives. So last year we had the pleasure of welcoming three new senior joiners in our team, they're coming from the [United] States, Germany and Australia. And actually they've built a very successful career with other quant shops. And the first task we gave to the three of them as new joiners was to roast the whole quant team after the first 100 days and say, “challenge us”. Challenging the team on things we do and what could have been done differently and what you find strange. Let’s debate and reflect on that. I have to say who that was, whew!

[Laughter]

EM: Was it a tough one?

WZ: Yes. Sometimes I was like sweating, like, oh, is it going the right direction? Exactly. But at the end of the day, it was so helpful that with all these outside-in, nice experiences, we reflected and we looked at the way we work and we improved a lot.

EM: So it's clear: it requires vision, it’s quite a big shift. We said it's the ‘next generation’. What about clients? Is it a process to take clients along with you in this shift?

WG: Yeah, we do a lot of research actually, in consultation with our clients. There's also because we think transparency is very important towards our clients, but also ourselves as portfolio managers. We think it's very important to understand why stocks are in the portfolio, for example. And so what we do in client meetings is that we share quite a lot of what we are working on, and also research that doesn't work. Dismissed research, as we like to call it, or research that we put aside for a while, and maybe we'll pick up later.

So we disclose quite a lot. We also write client notes. We write an academic papers. And yeah, because we think it's important to elaborate on why we also augment these more established, well-known factors with these new innovative signals. And maybe it's a nice quote from a recent client that mentioned [in their] feedback to us and said it seems that you are in the sweet spot between the use of more well-known, established factors and also these newer innovative signals. And I thought that was a big compliment, because indeed, that's also how I see that we work at Robeco.

EM: A very clear concluding message. Weili, thanks so much for your time and insights. Wilma, thank you for being with us; Iman, great to hear your perspectives and insights.

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.