I started at Robeco in 1995, joining the recently established quantitative research team. Back then the financial industry was still ruled by human decision-making, and Robeco’s investment strategies were all fully fundamentally managed. Smoking in the office was normal, and email was gradually being introduced.

My alma mater still offered a master’s in Japanese studies because the Japanese economic model was widely regarded as a success story that the West should emulate – although the Japanese equity and real estate bubble had already burst several years before. That university had also tried to convince me the market was efficient, with stock prices following the CAPM.

Our team was inspired by the emerging academic research challenging market efficiency. It showed that simple systematic strategies such as value and momentum could beat the market – at least on paper. We aimed to turn such insights into profitable real-life quantitative investment strategies. Back then, the US-centric literature only looked at single factors, and conveniently ignored matters such as transaction costs and benchmarks.

From the outset, we took a global perspective and found that not all results for US equities were equally robust when tested on international markets. We also learned how to combine different types of factors to maximize their diversification power, and what kind of buy and sell strategy to use to preserve the highest after-cost performance.

In a financial industry largely running on intuition and anecdotes, we became rational investors. Our premise was that although past returns are not a guarantee for the future, studying what did and didn’t work in the past, and understanding why should give investors an advantage. Our culture and investment philosophy was shaped during this period and in particular our evidence-based approach.

I’ve lost count of how many ideas intuitively looked promising but didn’t hold up when we put them to the test. Another principle that shaped us is that quantitative investment strategies should have a clear economic rationale. We are, after all, investors applying economic insights in a systematic manner, rather than mathematicians applying formulas to economic problems.

Our first quantitative equity strategies were launched in the early 2000s. These were enhanced indexing strategies, managed with a low tracking error against the preferred benchmark of our clients (typically a global or emerging markets index). Shortly after, we launched our Conservative Equities strategies, designed to benefit from the low volatility anomaly which was still fairly unknown then.

As low volatility investing comes with a high tracking error, it requires an absolute performance perspective instead of a benchmark relative approach, thus complementing our enhanced indexing strategies. For investors wanting the flexibility to make their own mix of exposures, we also made the individual factors available separately, turning them into building blocks for a tailor-made multi-factor portfolio.

Our strong track-records led to a tenfold increase in our quantitative assets under management over the past decade. For our research efforts, this means that transaction costs and capacity management have become increasingly important, especially since a sizable part of our assets under management are in emerging markets equities, where costs are relatively high and liquidity is generally lower. We also massively stepped our ESG research activities, as we aim to be leading with sustainability integration, and also because of increasing client demand for ESG customization.

Over the years, I remained aware of the risk of becoming complacent. I fostered a culture where even the most junior researcher was encouraged to challenge the status quo. I am happy that this is still around, 25 years later. To illustrate, you don’t get ahead in our team by merely doing exactly as you are told, but by being creative and innovative.

According to our HR department, personal KPIs should always be SMART, but my favorite KPI for researchers has always been: “Just surprise me!”. Given the complex nature of financial markets, there is no such thing as the perfect model, and since markets continuously evolve, so should quantitative investors.

Quantitative investment strategies have generally underperformed since the middle of 2018, with the underperformance being particularly severe in developed markets. Our long-term (>10 year) track-records are still strong. In fact, if I could have locked in these realized average returns when we started these strategies, I would have. But although our long-term average realized returns are on target, our more recent returns are clearly not.

The current drawdown is particularly painful in comparison to the relatively steady outperformance we realized before 2018. A drawdown of this magnitude and length in time is unique both to our live track records and our back tests covering many decades. I’m not surprised that some investors have started wondering whether quant investing might be impaired, or perhaps even permanently broken.

Before addressing this concern, let me elaborate briefly on why our strategies underperformed. The main cause of the current ‘quant winter’ is growth and large stocks massively outperforming their value and small counterparts, especially in developed markets, and within developed markets particularly in the US.

This has had a big impact because value is a key factor in all our models, and our quantitative strategies typically have a significant tilt towards small stocks. The latter is not because we especially believe in the small-cap premium – we don’t – but because quantitative strategies require a broad opportunity set and are generally more effective in the small-cap space.

Small stocks also have virtually zero weight in standard capitalization-weighted benchmark indices. The resulting overweight in value and smaller stocks is quite consistent across our flagship strategies, but also means these have all been somewhat affected by the underperformance of these factors, to a bigger or smaller extent.

The late nineties’ tech bubble saw a comparable meltdown of the value factor. But then, the large losses of the value factor were mirrored by similar-sized gains for the momentum factor, so a simple combination of the two sufficed for staying afloat during that period. Unfortunately, neither the momentum factor nor other factors, such as quality or low volatility, have been able to compensate for the recent value losses.

Although the other factors delivered positive returns or held their ground, they produced nowhere near the gains needed to offset the massive losses of the value factor. Diversifying across multiple uncorrelated factors usually offers the best protection against large losses on any single factor, but this time it has failed to prevent a sizable underperformance at the portfolio level. This explains why quant investors are experiencing a dark and cold winter.

The value factor’s misfortunes make one wonder whether the value premium has disappeared for good. In order to assess whether the value premium still exists, it is important to take a step back and consider why there has been a value premium in the first place. Some have argued that it reflects a risk premium, while others attribute it to systematic mispricing.

Is it a risk premium? This notion reflects an efficient market mindset: if something gives a higher return, then this must be a reward for some kind of risk. This would be good news for the value premium because risks don’t just vanish, so neither should their associated premiums. Unless one can come up with a very good explanation for why the risk that always needed to be rewarded with a premium suddenly doesn’t need to be rewarded anymore, the value premium should still exist.

However, I have always found the risk premium story rather opportunistic, because if growth stocks had outperformed value stocks historically, this growth premium would probably also be interpreted as a risk premium, even though it has the opposite sign. To make the argument stick, one has to show why value stocks are much riskier than growth stocks, or the other way around if the premium had the other sign! Empirically, however, both types of stocks appear to be equally risky, so the risk premium explanation seems implausible.

The main competing hypothesis is that the value premium stems from systematic mispricing. Unlike a risk premium, mispricing may be corrected if investors structurally adapt their behavior. Could the value factor for instance have been arbitraged away? Unlikely, because in that scenario one would expect to have seen a tremendous outperformance of value stocks relative to growth stocks, and a subsequent narrowing of valuation spreads between these two styles.

But instead of a value rally and a compression of multiple spreads, we have seen a spectacular growth rally resulting in a steep spread widening. In fact, the valuation spread between growth stocks and value stocks has reached record levels last seen in the late nineties, at the height of the tech bubble. This real-life observation is at odds with the concern that the value premium might have been arbitraged away.

The mispricing of securities is driven by investors’ behavioral biases, which is commonly interpreted as investors acting irrationally. I think this is a caricature. Behavioral biases do not necessarily imply irrationality, and the stakes in financial markets are too high to believe that investors keep making the same elementary mistakes in their decision making.

With hindsight, for example, it is easy to say that the dot.com bubble would inevitably burst. But at the time, there were many good reasons to believe otherwise. The emerging internet was disrupting business models, and the future appeared bleak for many ‘old economy’ stocks. Today there are again myriad narratives to justify the massive outperformance of growth stocks. Throughout the 20th century, there have been many extreme macro-economic circumstances, all of which the value factor survived.

I see the risk of investors eagerly embracing an ex-post rationalization appearing to justify what they see, which underestimates the resilience of the value factor. But we cannot be sure: maybe this time really is different. And precisely this doubt means it is not necessarily irrational to buy growth, even if current valuations appear excessive.

But if the behavioral biases that are the source of the value premium are not necessarily irrational, then what kind of rational drivers are we talking about? I think incentives are the key. Contrary to what economic models commonly assume, professional investors are not homo economicus, merely concerned with optimizing their long-term expected return-to-risk ratio.

They have career concerns. They deal with demanding bosses and clients with high expectations and are evaluated not only on long-term performance but also short-term results. Moreover, they must have an appealing story to market or explain their strategy; they must beat a benchmark or run a high-conviction portfolio containing only a few dozen names. And they want to sleep well at night and care about what their friends think.

In fact, if I were a fundamental instead of a quant investor, I can think of many good reasons to opt for a growth style – despite my solid belief in the value premium. If incentives are the source of the value premium, then it would be at risk if these fundamentally changed. However, incentives appear pretty much the same to me as they were five or ten years ago.

Having discussed risk-based versus mispricing-based explanations for the existence of factor premiums, I would like to argue that either way, painful drawdowns cannot be fully eliminated, but actually must occur occasionally for factor premiums to exist. In terms of risk-based explanations, this argument is straightforward.

Taking the equity risk premium – the mother of all risk premiums – as an example, it is widely accepted that this does not materialize in the form of gentle, steady outperformance of stocks over bonds every year, but comes with violent up- and downswings. Such is the nature of risk premiums.

Thus, if factor premiums are risk premiums, investors should expect major ups and downs rather than a steady return. Although diversifying across different factors might reduce this risk, it’s unrealistic to assume it can be eliminated entirely, just like the risk involved with the equity risk premium cannot be diversified to zero by investing in thousands of stocks.

If factor premiums are driven by mispricing, then occasional drawdowns are also part of the game. If, for instance, the value premium always manifested itself as a solid positive return, every investor would become a value investor, and the premium would not be sustainable.

Precisely because there are periods when value stocks suffer and growth stocks have a great run, many investors are lured into overpaying for growth and forgetting about value, which enforces and sustains the mispricing that is needed for a value premium to remain there – for those brave enough to stick to their guns.

Factors that have been around for a century are unlikely to have suddenly disappeared. It might be different if the rules of the game or financial incentives had transformed, but they haven’t. Risks that command a premium should continue to be rewarded with a premium, and mispricing resulting from investor behavior should be persistent as well, since our behavioral tendencies are pretty much hardwired in our DNA.

In fact, it seems that mispricing in the stock market has exacerbated in recent years. When valuation spreads were last at comparable levels, they were dubbed ‘irrational exuberance’. In light of current valuations, I would argue that the expected return on the value factor is currently well above its historical average.

I’m not predicting a performance reversal is imminent, because valuations could remain stretched or widen even further. As John Maynard Keynes used to say, markets can stay irrational longer than investors can stay solvent. No one can say if or when the big turnaround will come, but one thing is clear: the long-term expected return on the value factor currently looks brighter than ever.

Expected returns for multi-factor strategies have also improved, now that there is again some healthy quant investing skepticism among investors and crowding has become less of a concern. After all, strategies shouldn’t become too popular for their own good.

This does not mean that at Robeco’s quant group we are calmly sitting this winter out, waiting for spring to come. On the contrary. We are not amused, to put it mildly, by the big dent in our track records and we share – as fellow investors in our strategies – the pain experienced by our clients. This drawdown has led us to go back to the drawing board and review every step and decision in our investment process.

Have we perhaps had too much value and small-cap exposure in our portfolios? Performance in recent years would clearly have been better with less exposure to these two now hammered factors. In other periods, however, these same factors made a strong positive contribution to our performance.

Moreover, other factors have also experienced crashes, such as momentum in 2009. Thus, the challenge is to find an optimal mix of different factors that best prepares us for the decades ahead, taking into account both past scenarios and potentially very different future ones.

As well as reassessing our factor composition, we are examining whether we can further enhance their definitions. Looking at the value factor again as an example, are classic metrics such as the ratio of book value to market value still optimal? Book values might be informative for traditional industrial firms, with factories and machines, but are arguably less relevant for service-oriented firms such as Facebook and Netflix.

Here, value metrics could be enhanced by augmenting the fundamentals reported in financial statements with intangible assets such as knowledge capital, brand value, and human capital. The challenge, of course, is to quantify these intangibles.

Another research angle is to de-risk the value factor by looking more at the relative value of a stock compared to its most relevant peers. For instance, whereas a traditional value strategy would not buy any FANMAG stocks because they are all expensive, one could also choose to compare FANMAG(-like) stocks with each other and look for relative value among peers.

Only so much can be squeezed out of traditional factors, however. Ultimately, we need a richer set of uncorrelated factors: fresh alpha sources. Academic literature, for example, provides a ‘factor zoo’ consisting of hundreds of factors, although many of these factors are variations on the same theme, using the same data in slightly different ways.

Alternative datasets, which are rapidly becoming available nowadays as part of the broader big data revolution, are perhaps more promising. These might also require new processing techniques such as machine learning algorithms since the data they contain might be unstructured or textual instead of numerical. With more than a thousand alternative datasets already available, quantitative investing is ready to be taken to the next level. Right now, this is our top research priority.

In my career as a quant investor, I’ve experienced the occasional cooler quant season in which factors struggled to deliver. Although this feels for most of us like a long and dark winter, I am more optimistic than ever that this period of underperformance will be followed by a bright new era.

I recognize this current period raises questions about the future viability of long-running strategies, but experience has taught me that the darkest hour is just before dawn. We will notice spring’s arrival not only by a better performance of our strategies but also by the birth of new strategy enhancements that will increase our resilience in future quant seasons to come.